1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2 ; RUN: opt < %s -O3 -mcpu=knl -S | FileCheck %s -check-prefix=AVX512
3 ; RUN: opt < %s -O3 -mcpu=knl -force-vector-width=2 -S | FileCheck %s -check-prefix=FVW2
5 ; With a force-vector-width, it is sometimes more profitable to generate
6 ; scalarized and predicated stores instead of masked scatter.
8 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
9 target triple = "x86_64-pc_linux"
13 ;void foo1(float * __restrict__ in, float * __restrict__ out, int * __restrict__ trigger, int * __restrict__ index) {
15 ; for (int i=0; i < SIZE; ++i) {
16 ; if (trigger[i] > 0) {
17 ; out[i] = in[index[i]] + (float) 0.5;
22 ; Function Attrs: nounwind uwtable
23 define void @foo1(float* noalias %in, float* noalias %out, i32* noalias %trigger, i32* noalias %index) {
24 ; AVX512-LABEL: @foo1(
25 ; AVX512-NEXT: iter.check:
26 ; AVX512-NEXT: br label [[VECTOR_BODY:%.*]]
27 ; AVX512: vector.body:
28 ; AVX512-NEXT: [[INDEX6:%.*]] = phi i64 [ 0, [[ITER_CHECK:%.*]] ], [ [[INDEX_NEXT_3:%.*]], [[VECTOR_BODY]] ]
29 ; AVX512-NEXT: [[TMP0:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER:%.*]], i64 [[INDEX6]]
30 ; AVX512-NEXT: [[TMP1:%.*]] = bitcast i32* [[TMP0]] to <16 x i32>*
31 ; AVX512-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i32>, <16 x i32>* [[TMP1]], align 4
32 ; AVX512-NEXT: [[TMP2:%.*]] = icmp sgt <16 x i32> [[WIDE_LOAD]], zeroinitializer
33 ; AVX512-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, i32* [[INDEX:%.*]], i64 [[INDEX6]]
34 ; AVX512-NEXT: [[TMP4:%.*]] = bitcast i32* [[TMP3]] to <16 x i32>*
35 ; AVX512-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call <16 x i32> @llvm.masked.load.v16i32.p0v16i32(<16 x i32>* [[TMP4]], i32 4, <16 x i1> [[TMP2]], <16 x i32> poison)
36 ; AVX512-NEXT: [[TMP5:%.*]] = sext <16 x i32> [[WIDE_MASKED_LOAD]] to <16 x i64>
37 ; AVX512-NEXT: [[TMP6:%.*]] = getelementptr inbounds float, float* [[IN:%.*]], <16 x i64> [[TMP5]]
38 ; AVX512-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP6]], i32 4, <16 x i1> [[TMP2]], <16 x float> undef)
39 ; AVX512-NEXT: [[TMP7:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
40 ; AVX512-NEXT: [[TMP8:%.*]] = getelementptr inbounds float, float* [[OUT:%.*]], i64 [[INDEX6]]
41 ; AVX512-NEXT: [[TMP9:%.*]] = bitcast float* [[TMP8]] to <16 x float>*
42 ; AVX512-NEXT: call void @llvm.masked.store.v16f32.p0v16f32(<16 x float> [[TMP7]], <16 x float>* [[TMP9]], i32 4, <16 x i1> [[TMP2]])
43 ; AVX512-NEXT: [[INDEX_NEXT:%.*]] = or i64 [[INDEX6]], 16
44 ; AVX512-NEXT: [[TMP10:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], i64 [[INDEX_NEXT]]
45 ; AVX512-NEXT: [[TMP11:%.*]] = bitcast i32* [[TMP10]] to <16 x i32>*
46 ; AVX512-NEXT: [[WIDE_LOAD_1:%.*]] = load <16 x i32>, <16 x i32>* [[TMP11]], align 4
47 ; AVX512-NEXT: [[TMP12:%.*]] = icmp sgt <16 x i32> [[WIDE_LOAD_1]], zeroinitializer
48 ; AVX512-NEXT: [[TMP13:%.*]] = getelementptr inbounds i32, i32* [[INDEX]], i64 [[INDEX_NEXT]]
49 ; AVX512-NEXT: [[TMP14:%.*]] = bitcast i32* [[TMP13]] to <16 x i32>*
50 ; AVX512-NEXT: [[WIDE_MASKED_LOAD_1:%.*]] = call <16 x i32> @llvm.masked.load.v16i32.p0v16i32(<16 x i32>* nonnull [[TMP14]], i32 4, <16 x i1> [[TMP12]], <16 x i32> poison)
51 ; AVX512-NEXT: [[TMP15:%.*]] = sext <16 x i32> [[WIDE_MASKED_LOAD_1]] to <16 x i64>
52 ; AVX512-NEXT: [[TMP16:%.*]] = getelementptr inbounds float, float* [[IN]], <16 x i64> [[TMP15]]
53 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_1:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP16]], i32 4, <16 x i1> [[TMP12]], <16 x float> undef)
54 ; AVX512-NEXT: [[TMP17:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER_1]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
55 ; AVX512-NEXT: [[TMP18:%.*]] = getelementptr inbounds float, float* [[OUT]], i64 [[INDEX_NEXT]]
56 ; AVX512-NEXT: [[TMP19:%.*]] = bitcast float* [[TMP18]] to <16 x float>*
57 ; AVX512-NEXT: call void @llvm.masked.store.v16f32.p0v16f32(<16 x float> [[TMP17]], <16 x float>* [[TMP19]], i32 4, <16 x i1> [[TMP12]])
58 ; AVX512-NEXT: [[INDEX_NEXT_1:%.*]] = or i64 [[INDEX6]], 32
59 ; AVX512-NEXT: [[TMP20:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], i64 [[INDEX_NEXT_1]]
60 ; AVX512-NEXT: [[TMP21:%.*]] = bitcast i32* [[TMP20]] to <16 x i32>*
61 ; AVX512-NEXT: [[WIDE_LOAD_2:%.*]] = load <16 x i32>, <16 x i32>* [[TMP21]], align 4
62 ; AVX512-NEXT: [[TMP22:%.*]] = icmp sgt <16 x i32> [[WIDE_LOAD_2]], zeroinitializer
63 ; AVX512-NEXT: [[TMP23:%.*]] = getelementptr inbounds i32, i32* [[INDEX]], i64 [[INDEX_NEXT_1]]
64 ; AVX512-NEXT: [[TMP24:%.*]] = bitcast i32* [[TMP23]] to <16 x i32>*
65 ; AVX512-NEXT: [[WIDE_MASKED_LOAD_2:%.*]] = call <16 x i32> @llvm.masked.load.v16i32.p0v16i32(<16 x i32>* nonnull [[TMP24]], i32 4, <16 x i1> [[TMP22]], <16 x i32> poison)
66 ; AVX512-NEXT: [[TMP25:%.*]] = sext <16 x i32> [[WIDE_MASKED_LOAD_2]] to <16 x i64>
67 ; AVX512-NEXT: [[TMP26:%.*]] = getelementptr inbounds float, float* [[IN]], <16 x i64> [[TMP25]]
68 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_2:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP26]], i32 4, <16 x i1> [[TMP22]], <16 x float> undef)
69 ; AVX512-NEXT: [[TMP27:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER_2]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
70 ; AVX512-NEXT: [[TMP28:%.*]] = getelementptr inbounds float, float* [[OUT]], i64 [[INDEX_NEXT_1]]
71 ; AVX512-NEXT: [[TMP29:%.*]] = bitcast float* [[TMP28]] to <16 x float>*
72 ; AVX512-NEXT: call void @llvm.masked.store.v16f32.p0v16f32(<16 x float> [[TMP27]], <16 x float>* [[TMP29]], i32 4, <16 x i1> [[TMP22]])
73 ; AVX512-NEXT: [[INDEX_NEXT_2:%.*]] = or i64 [[INDEX6]], 48
74 ; AVX512-NEXT: [[TMP30:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], i64 [[INDEX_NEXT_2]]
75 ; AVX512-NEXT: [[TMP31:%.*]] = bitcast i32* [[TMP30]] to <16 x i32>*
76 ; AVX512-NEXT: [[WIDE_LOAD_3:%.*]] = load <16 x i32>, <16 x i32>* [[TMP31]], align 4
77 ; AVX512-NEXT: [[TMP32:%.*]] = icmp sgt <16 x i32> [[WIDE_LOAD_3]], zeroinitializer
78 ; AVX512-NEXT: [[TMP33:%.*]] = getelementptr inbounds i32, i32* [[INDEX]], i64 [[INDEX_NEXT_2]]
79 ; AVX512-NEXT: [[TMP34:%.*]] = bitcast i32* [[TMP33]] to <16 x i32>*
80 ; AVX512-NEXT: [[WIDE_MASKED_LOAD_3:%.*]] = call <16 x i32> @llvm.masked.load.v16i32.p0v16i32(<16 x i32>* nonnull [[TMP34]], i32 4, <16 x i1> [[TMP32]], <16 x i32> poison)
81 ; AVX512-NEXT: [[TMP35:%.*]] = sext <16 x i32> [[WIDE_MASKED_LOAD_3]] to <16 x i64>
82 ; AVX512-NEXT: [[TMP36:%.*]] = getelementptr inbounds float, float* [[IN]], <16 x i64> [[TMP35]]
83 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_3:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP36]], i32 4, <16 x i1> [[TMP32]], <16 x float> undef)
84 ; AVX512-NEXT: [[TMP37:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER_3]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
85 ; AVX512-NEXT: [[TMP38:%.*]] = getelementptr inbounds float, float* [[OUT]], i64 [[INDEX_NEXT_2]]
86 ; AVX512-NEXT: [[TMP39:%.*]] = bitcast float* [[TMP38]] to <16 x float>*
87 ; AVX512-NEXT: call void @llvm.masked.store.v16f32.p0v16f32(<16 x float> [[TMP37]], <16 x float>* [[TMP39]], i32 4, <16 x i1> [[TMP32]])
88 ; AVX512-NEXT: [[INDEX_NEXT_3]] = add nuw nsw i64 [[INDEX6]], 64
89 ; AVX512-NEXT: [[TMP40:%.*]] = icmp eq i64 [[INDEX_NEXT_3]], 4096
90 ; AVX512-NEXT: br i1 [[TMP40]], label [[FOR_END:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
92 ; AVX512-NEXT: ret void
96 ; FVW2-NEXT: br label [[VECTOR_BODY:%.*]]
98 ; FVW2-NEXT: [[INDEX6:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
99 ; FVW2-NEXT: [[TMP0:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER:%.*]], i64 [[INDEX6]]
100 ; FVW2-NEXT: [[TMP1:%.*]] = bitcast i32* [[TMP0]] to <2 x i32>*
101 ; FVW2-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i32>, <2 x i32>* [[TMP1]], align 4
102 ; FVW2-NEXT: [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[TMP0]], i64 2
103 ; FVW2-NEXT: [[TMP3:%.*]] = bitcast i32* [[TMP2]] to <2 x i32>*
104 ; FVW2-NEXT: [[WIDE_LOAD7:%.*]] = load <2 x i32>, <2 x i32>* [[TMP3]], align 4
105 ; FVW2-NEXT: [[TMP4:%.*]] = getelementptr inbounds i32, i32* [[TMP0]], i64 4
106 ; FVW2-NEXT: [[TMP5:%.*]] = bitcast i32* [[TMP4]] to <2 x i32>*
107 ; FVW2-NEXT: [[WIDE_LOAD8:%.*]] = load <2 x i32>, <2 x i32>* [[TMP5]], align 4
108 ; FVW2-NEXT: [[TMP6:%.*]] = getelementptr inbounds i32, i32* [[TMP0]], i64 6
109 ; FVW2-NEXT: [[TMP7:%.*]] = bitcast i32* [[TMP6]] to <2 x i32>*
110 ; FVW2-NEXT: [[WIDE_LOAD9:%.*]] = load <2 x i32>, <2 x i32>* [[TMP7]], align 4
111 ; FVW2-NEXT: [[TMP8:%.*]] = icmp sgt <2 x i32> [[WIDE_LOAD]], zeroinitializer
112 ; FVW2-NEXT: [[TMP9:%.*]] = icmp sgt <2 x i32> [[WIDE_LOAD7]], zeroinitializer
113 ; FVW2-NEXT: [[TMP10:%.*]] = icmp sgt <2 x i32> [[WIDE_LOAD8]], zeroinitializer
114 ; FVW2-NEXT: [[TMP11:%.*]] = icmp sgt <2 x i32> [[WIDE_LOAD9]], zeroinitializer
115 ; FVW2-NEXT: [[TMP12:%.*]] = getelementptr inbounds i32, i32* [[INDEX:%.*]], i64 [[INDEX6]]
116 ; FVW2-NEXT: [[TMP13:%.*]] = bitcast i32* [[TMP12]] to <2 x i32>*
117 ; FVW2-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call <2 x i32> @llvm.masked.load.v2i32.p0v2i32(<2 x i32>* [[TMP13]], i32 4, <2 x i1> [[TMP8]], <2 x i32> poison)
118 ; FVW2-NEXT: [[TMP14:%.*]] = getelementptr inbounds i32, i32* [[TMP12]], i64 2
119 ; FVW2-NEXT: [[TMP15:%.*]] = bitcast i32* [[TMP14]] to <2 x i32>*
120 ; FVW2-NEXT: [[WIDE_MASKED_LOAD10:%.*]] = call <2 x i32> @llvm.masked.load.v2i32.p0v2i32(<2 x i32>* nonnull [[TMP15]], i32 4, <2 x i1> [[TMP9]], <2 x i32> poison)
121 ; FVW2-NEXT: [[TMP16:%.*]] = getelementptr inbounds i32, i32* [[TMP12]], i64 4
122 ; FVW2-NEXT: [[TMP17:%.*]] = bitcast i32* [[TMP16]] to <2 x i32>*
123 ; FVW2-NEXT: [[WIDE_MASKED_LOAD11:%.*]] = call <2 x i32> @llvm.masked.load.v2i32.p0v2i32(<2 x i32>* nonnull [[TMP17]], i32 4, <2 x i1> [[TMP10]], <2 x i32> poison)
124 ; FVW2-NEXT: [[TMP18:%.*]] = getelementptr inbounds i32, i32* [[TMP12]], i64 6
125 ; FVW2-NEXT: [[TMP19:%.*]] = bitcast i32* [[TMP18]] to <2 x i32>*
126 ; FVW2-NEXT: [[WIDE_MASKED_LOAD12:%.*]] = call <2 x i32> @llvm.masked.load.v2i32.p0v2i32(<2 x i32>* nonnull [[TMP19]], i32 4, <2 x i1> [[TMP11]], <2 x i32> poison)
127 ; FVW2-NEXT: [[TMP20:%.*]] = sext <2 x i32> [[WIDE_MASKED_LOAD]] to <2 x i64>
128 ; FVW2-NEXT: [[TMP21:%.*]] = sext <2 x i32> [[WIDE_MASKED_LOAD10]] to <2 x i64>
129 ; FVW2-NEXT: [[TMP22:%.*]] = sext <2 x i32> [[WIDE_MASKED_LOAD11]] to <2 x i64>
130 ; FVW2-NEXT: [[TMP23:%.*]] = sext <2 x i32> [[WIDE_MASKED_LOAD12]] to <2 x i64>
131 ; FVW2-NEXT: [[TMP24:%.*]] = getelementptr inbounds float, float* [[IN:%.*]], <2 x i64> [[TMP20]]
132 ; FVW2-NEXT: [[TMP25:%.*]] = getelementptr inbounds float, float* [[IN]], <2 x i64> [[TMP21]]
133 ; FVW2-NEXT: [[TMP26:%.*]] = getelementptr inbounds float, float* [[IN]], <2 x i64> [[TMP22]]
134 ; FVW2-NEXT: [[TMP27:%.*]] = getelementptr inbounds float, float* [[IN]], <2 x i64> [[TMP23]]
135 ; FVW2-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <2 x float> @llvm.masked.gather.v2f32.v2p0f32(<2 x float*> [[TMP24]], i32 4, <2 x i1> [[TMP8]], <2 x float> undef)
136 ; FVW2-NEXT: [[WIDE_MASKED_GATHER13:%.*]] = call <2 x float> @llvm.masked.gather.v2f32.v2p0f32(<2 x float*> [[TMP25]], i32 4, <2 x i1> [[TMP9]], <2 x float> undef)
137 ; FVW2-NEXT: [[WIDE_MASKED_GATHER14:%.*]] = call <2 x float> @llvm.masked.gather.v2f32.v2p0f32(<2 x float*> [[TMP26]], i32 4, <2 x i1> [[TMP10]], <2 x float> undef)
138 ; FVW2-NEXT: [[WIDE_MASKED_GATHER15:%.*]] = call <2 x float> @llvm.masked.gather.v2f32.v2p0f32(<2 x float*> [[TMP27]], i32 4, <2 x i1> [[TMP11]], <2 x float> undef)
139 ; FVW2-NEXT: [[TMP28:%.*]] = fadd <2 x float> [[WIDE_MASKED_GATHER]], <float 5.000000e-01, float 5.000000e-01>
140 ; FVW2-NEXT: [[TMP29:%.*]] = fadd <2 x float> [[WIDE_MASKED_GATHER13]], <float 5.000000e-01, float 5.000000e-01>
141 ; FVW2-NEXT: [[TMP30:%.*]] = fadd <2 x float> [[WIDE_MASKED_GATHER14]], <float 5.000000e-01, float 5.000000e-01>
142 ; FVW2-NEXT: [[TMP31:%.*]] = fadd <2 x float> [[WIDE_MASKED_GATHER15]], <float 5.000000e-01, float 5.000000e-01>
143 ; FVW2-NEXT: [[TMP32:%.*]] = getelementptr inbounds float, float* [[OUT:%.*]], i64 [[INDEX6]]
144 ; FVW2-NEXT: [[TMP33:%.*]] = bitcast float* [[TMP32]] to <2 x float>*
145 ; FVW2-NEXT: call void @llvm.masked.store.v2f32.p0v2f32(<2 x float> [[TMP28]], <2 x float>* [[TMP33]], i32 4, <2 x i1> [[TMP8]])
146 ; FVW2-NEXT: [[TMP34:%.*]] = getelementptr inbounds float, float* [[TMP32]], i64 2
147 ; FVW2-NEXT: [[TMP35:%.*]] = bitcast float* [[TMP34]] to <2 x float>*
148 ; FVW2-NEXT: call void @llvm.masked.store.v2f32.p0v2f32(<2 x float> [[TMP29]], <2 x float>* [[TMP35]], i32 4, <2 x i1> [[TMP9]])
149 ; FVW2-NEXT: [[TMP36:%.*]] = getelementptr inbounds float, float* [[TMP32]], i64 4
150 ; FVW2-NEXT: [[TMP37:%.*]] = bitcast float* [[TMP36]] to <2 x float>*
151 ; FVW2-NEXT: call void @llvm.masked.store.v2f32.p0v2f32(<2 x float> [[TMP30]], <2 x float>* [[TMP37]], i32 4, <2 x i1> [[TMP10]])
152 ; FVW2-NEXT: [[TMP38:%.*]] = getelementptr inbounds float, float* [[TMP32]], i64 6
153 ; FVW2-NEXT: [[TMP39:%.*]] = bitcast float* [[TMP38]] to <2 x float>*
154 ; FVW2-NEXT: call void @llvm.masked.store.v2f32.p0v2f32(<2 x float> [[TMP31]], <2 x float>* [[TMP39]], i32 4, <2 x i1> [[TMP11]])
155 ; FVW2-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX6]], 8
156 ; FVW2-NEXT: [[TMP40:%.*]] = icmp eq i64 [[INDEX_NEXT]], 4096
157 ; FVW2-NEXT: br i1 [[TMP40]], label [[FOR_END:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
159 ; FVW2-NEXT: ret void
162 %in.addr = alloca float*, align 8
163 %out.addr = alloca float*, align 8
164 %trigger.addr = alloca i32*, align 8
165 %index.addr = alloca i32*, align 8
166 %i = alloca i32, align 4
167 store float* %in, float** %in.addr, align 8
168 store float* %out, float** %out.addr, align 8
169 store i32* %trigger, i32** %trigger.addr, align 8
170 store i32* %index, i32** %index.addr, align 8
171 store i32 0, i32* %i, align 4
174 for.cond: ; preds = %for.inc, %entry
175 %0 = load i32, i32* %i, align 4
176 %cmp = icmp slt i32 %0, 4096
177 br i1 %cmp, label %for.body, label %for.end
179 for.body: ; preds = %for.cond
180 %1 = load i32, i32* %i, align 4
181 %idxprom = sext i32 %1 to i64
182 %2 = load i32*, i32** %trigger.addr, align 8
183 %arrayidx = getelementptr inbounds i32, i32* %2, i64 %idxprom
184 %3 = load i32, i32* %arrayidx, align 4
185 %cmp1 = icmp sgt i32 %3, 0
186 br i1 %cmp1, label %if.then, label %if.end
188 if.then: ; preds = %for.body
189 %4 = load i32, i32* %i, align 4
190 %idxprom2 = sext i32 %4 to i64
191 %5 = load i32*, i32** %index.addr, align 8
192 %arrayidx3 = getelementptr inbounds i32, i32* %5, i64 %idxprom2
193 %6 = load i32, i32* %arrayidx3, align 4
194 %idxprom4 = sext i32 %6 to i64
195 %7 = load float*, float** %in.addr, align 8
196 %arrayidx5 = getelementptr inbounds float, float* %7, i64 %idxprom4
197 %8 = load float, float* %arrayidx5, align 4
198 %add = fadd float %8, 5.000000e-01
199 %9 = load i32, i32* %i, align 4
200 %idxprom6 = sext i32 %9 to i64
201 %10 = load float*, float** %out.addr, align 8
202 %arrayidx7 = getelementptr inbounds float, float* %10, i64 %idxprom6
203 store float %add, float* %arrayidx7, align 4
206 if.end: ; preds = %if.then, %for.body
209 for.inc: ; preds = %if.end
210 %11 = load i32, i32* %i, align 4
211 %inc = add nsw i32 %11, 1
212 store i32 %inc, i32* %i, align 4
215 for.end: ; preds = %for.cond
220 ;void foo2 (In * __restrict__ in, float * __restrict__ out, int * __restrict__ trigger) {
222 ; for (int i=0; i<SIZE; i += 16) {
223 ; if (trigger[i] > 0) {
224 ; out[i] = in[i].b + (float) 0.5;
229 %struct.In = type { float, float }
231 define void @foo2(%struct.In* noalias %in, float* noalias %out, i32* noalias %trigger, i32* noalias %index) #0 {
232 ; AVX512-LABEL: @foo2(
233 ; AVX512-NEXT: entry:
234 ; AVX512-NEXT: [[TMP0:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER:%.*]], <16 x i64> <i64 0, i64 16, i64 32, i64 48, i64 64, i64 80, i64 96, i64 112, i64 128, i64 144, i64 160, i64 176, i64 192, i64 208, i64 224, i64 240>
235 ; AVX512-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP0]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
236 ; AVX512-NEXT: [[TMP1:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER]], zeroinitializer
237 ; AVX512-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT_IN:%.*]], %struct.In* [[IN:%.*]], <16 x i64> <i64 0, i64 16, i64 32, i64 48, i64 64, i64 80, i64 96, i64 112, i64 128, i64 144, i64 160, i64 176, i64 192, i64 208, i64 224, i64 240>, i32 1
238 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP2]], i32 4, <16 x i1> [[TMP1]], <16 x float> undef)
239 ; AVX512-NEXT: [[TMP3:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
240 ; AVX512-NEXT: [[TMP4:%.*]] = getelementptr inbounds float, float* [[OUT:%.*]], <16 x i64> <i64 0, i64 16, i64 32, i64 48, i64 64, i64 80, i64 96, i64 112, i64 128, i64 144, i64 160, i64 176, i64 192, i64 208, i64 224, i64 240>
241 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP3]], <16 x float*> [[TMP4]], i32 4, <16 x i1> [[TMP1]])
242 ; AVX512-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 256, i64 272, i64 288, i64 304, i64 320, i64 336, i64 352, i64 368, i64 384, i64 400, i64 416, i64 432, i64 448, i64 464, i64 480, i64 496>
243 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_1:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP5]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
244 ; AVX512-NEXT: [[TMP6:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_1]], zeroinitializer
245 ; AVX512-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 256, i64 272, i64 288, i64 304, i64 320, i64 336, i64 352, i64 368, i64 384, i64 400, i64 416, i64 432, i64 448, i64 464, i64 480, i64 496>, i32 1
246 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_1:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP7]], i32 4, <16 x i1> [[TMP6]], <16 x float> undef)
247 ; AVX512-NEXT: [[TMP8:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_1]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
248 ; AVX512-NEXT: [[TMP9:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 256, i64 272, i64 288, i64 304, i64 320, i64 336, i64 352, i64 368, i64 384, i64 400, i64 416, i64 432, i64 448, i64 464, i64 480, i64 496>
249 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP8]], <16 x float*> [[TMP9]], i32 4, <16 x i1> [[TMP6]])
250 ; AVX512-NEXT: [[TMP10:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 512, i64 528, i64 544, i64 560, i64 576, i64 592, i64 608, i64 624, i64 640, i64 656, i64 672, i64 688, i64 704, i64 720, i64 736, i64 752>
251 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_2:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP10]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
252 ; AVX512-NEXT: [[TMP11:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_2]], zeroinitializer
253 ; AVX512-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 512, i64 528, i64 544, i64 560, i64 576, i64 592, i64 608, i64 624, i64 640, i64 656, i64 672, i64 688, i64 704, i64 720, i64 736, i64 752>, i32 1
254 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_2:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP12]], i32 4, <16 x i1> [[TMP11]], <16 x float> undef)
255 ; AVX512-NEXT: [[TMP13:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_2]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
256 ; AVX512-NEXT: [[TMP14:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 512, i64 528, i64 544, i64 560, i64 576, i64 592, i64 608, i64 624, i64 640, i64 656, i64 672, i64 688, i64 704, i64 720, i64 736, i64 752>
257 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP13]], <16 x float*> [[TMP14]], i32 4, <16 x i1> [[TMP11]])
258 ; AVX512-NEXT: [[TMP15:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 768, i64 784, i64 800, i64 816, i64 832, i64 848, i64 864, i64 880, i64 896, i64 912, i64 928, i64 944, i64 960, i64 976, i64 992, i64 1008>
259 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_3:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP15]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
260 ; AVX512-NEXT: [[TMP16:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_3]], zeroinitializer
261 ; AVX512-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 768, i64 784, i64 800, i64 816, i64 832, i64 848, i64 864, i64 880, i64 896, i64 912, i64 928, i64 944, i64 960, i64 976, i64 992, i64 1008>, i32 1
262 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_3:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP17]], i32 4, <16 x i1> [[TMP16]], <16 x float> undef)
263 ; AVX512-NEXT: [[TMP18:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_3]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
264 ; AVX512-NEXT: [[TMP19:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 768, i64 784, i64 800, i64 816, i64 832, i64 848, i64 864, i64 880, i64 896, i64 912, i64 928, i64 944, i64 960, i64 976, i64 992, i64 1008>
265 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP18]], <16 x float*> [[TMP19]], i32 4, <16 x i1> [[TMP16]])
266 ; AVX512-NEXT: [[TMP20:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 1024, i64 1040, i64 1056, i64 1072, i64 1088, i64 1104, i64 1120, i64 1136, i64 1152, i64 1168, i64 1184, i64 1200, i64 1216, i64 1232, i64 1248, i64 1264>
267 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_4:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP20]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
268 ; AVX512-NEXT: [[TMP21:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_4]], zeroinitializer
269 ; AVX512-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 1024, i64 1040, i64 1056, i64 1072, i64 1088, i64 1104, i64 1120, i64 1136, i64 1152, i64 1168, i64 1184, i64 1200, i64 1216, i64 1232, i64 1248, i64 1264>, i32 1
270 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_4:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP22]], i32 4, <16 x i1> [[TMP21]], <16 x float> undef)
271 ; AVX512-NEXT: [[TMP23:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_4]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
272 ; AVX512-NEXT: [[TMP24:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 1024, i64 1040, i64 1056, i64 1072, i64 1088, i64 1104, i64 1120, i64 1136, i64 1152, i64 1168, i64 1184, i64 1200, i64 1216, i64 1232, i64 1248, i64 1264>
273 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP23]], <16 x float*> [[TMP24]], i32 4, <16 x i1> [[TMP21]])
274 ; AVX512-NEXT: [[TMP25:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 1280, i64 1296, i64 1312, i64 1328, i64 1344, i64 1360, i64 1376, i64 1392, i64 1408, i64 1424, i64 1440, i64 1456, i64 1472, i64 1488, i64 1504, i64 1520>
275 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_5:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP25]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
276 ; AVX512-NEXT: [[TMP26:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_5]], zeroinitializer
277 ; AVX512-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 1280, i64 1296, i64 1312, i64 1328, i64 1344, i64 1360, i64 1376, i64 1392, i64 1408, i64 1424, i64 1440, i64 1456, i64 1472, i64 1488, i64 1504, i64 1520>, i32 1
278 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_5:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP27]], i32 4, <16 x i1> [[TMP26]], <16 x float> undef)
279 ; AVX512-NEXT: [[TMP28:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_5]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
280 ; AVX512-NEXT: [[TMP29:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 1280, i64 1296, i64 1312, i64 1328, i64 1344, i64 1360, i64 1376, i64 1392, i64 1408, i64 1424, i64 1440, i64 1456, i64 1472, i64 1488, i64 1504, i64 1520>
281 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP28]], <16 x float*> [[TMP29]], i32 4, <16 x i1> [[TMP26]])
282 ; AVX512-NEXT: [[TMP30:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 1536, i64 1552, i64 1568, i64 1584, i64 1600, i64 1616, i64 1632, i64 1648, i64 1664, i64 1680, i64 1696, i64 1712, i64 1728, i64 1744, i64 1760, i64 1776>
283 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_6:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP30]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
284 ; AVX512-NEXT: [[TMP31:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_6]], zeroinitializer
285 ; AVX512-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 1536, i64 1552, i64 1568, i64 1584, i64 1600, i64 1616, i64 1632, i64 1648, i64 1664, i64 1680, i64 1696, i64 1712, i64 1728, i64 1744, i64 1760, i64 1776>, i32 1
286 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_6:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP32]], i32 4, <16 x i1> [[TMP31]], <16 x float> undef)
287 ; AVX512-NEXT: [[TMP33:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_6]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
288 ; AVX512-NEXT: [[TMP34:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 1536, i64 1552, i64 1568, i64 1584, i64 1600, i64 1616, i64 1632, i64 1648, i64 1664, i64 1680, i64 1696, i64 1712, i64 1728, i64 1744, i64 1760, i64 1776>
289 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP33]], <16 x float*> [[TMP34]], i32 4, <16 x i1> [[TMP31]])
290 ; AVX512-NEXT: [[TMP35:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 1792, i64 1808, i64 1824, i64 1840, i64 1856, i64 1872, i64 1888, i64 1904, i64 1920, i64 1936, i64 1952, i64 1968, i64 1984, i64 2000, i64 2016, i64 2032>
291 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_7:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP35]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
292 ; AVX512-NEXT: [[TMP36:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_7]], zeroinitializer
293 ; AVX512-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 1792, i64 1808, i64 1824, i64 1840, i64 1856, i64 1872, i64 1888, i64 1904, i64 1920, i64 1936, i64 1952, i64 1968, i64 1984, i64 2000, i64 2016, i64 2032>, i32 1
294 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_7:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP37]], i32 4, <16 x i1> [[TMP36]], <16 x float> undef)
295 ; AVX512-NEXT: [[TMP38:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_7]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
296 ; AVX512-NEXT: [[TMP39:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 1792, i64 1808, i64 1824, i64 1840, i64 1856, i64 1872, i64 1888, i64 1904, i64 1920, i64 1936, i64 1952, i64 1968, i64 1984, i64 2000, i64 2016, i64 2032>
297 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP38]], <16 x float*> [[TMP39]], i32 4, <16 x i1> [[TMP36]])
298 ; AVX512-NEXT: [[TMP40:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 2048, i64 2064, i64 2080, i64 2096, i64 2112, i64 2128, i64 2144, i64 2160, i64 2176, i64 2192, i64 2208, i64 2224, i64 2240, i64 2256, i64 2272, i64 2288>
299 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_8:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP40]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
300 ; AVX512-NEXT: [[TMP41:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_8]], zeroinitializer
301 ; AVX512-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 2048, i64 2064, i64 2080, i64 2096, i64 2112, i64 2128, i64 2144, i64 2160, i64 2176, i64 2192, i64 2208, i64 2224, i64 2240, i64 2256, i64 2272, i64 2288>, i32 1
302 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_8:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP42]], i32 4, <16 x i1> [[TMP41]], <16 x float> undef)
303 ; AVX512-NEXT: [[TMP43:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_8]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
304 ; AVX512-NEXT: [[TMP44:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 2048, i64 2064, i64 2080, i64 2096, i64 2112, i64 2128, i64 2144, i64 2160, i64 2176, i64 2192, i64 2208, i64 2224, i64 2240, i64 2256, i64 2272, i64 2288>
305 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP43]], <16 x float*> [[TMP44]], i32 4, <16 x i1> [[TMP41]])
306 ; AVX512-NEXT: [[TMP45:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 2304, i64 2320, i64 2336, i64 2352, i64 2368, i64 2384, i64 2400, i64 2416, i64 2432, i64 2448, i64 2464, i64 2480, i64 2496, i64 2512, i64 2528, i64 2544>
307 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_9:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP45]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
308 ; AVX512-NEXT: [[TMP46:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_9]], zeroinitializer
309 ; AVX512-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 2304, i64 2320, i64 2336, i64 2352, i64 2368, i64 2384, i64 2400, i64 2416, i64 2432, i64 2448, i64 2464, i64 2480, i64 2496, i64 2512, i64 2528, i64 2544>, i32 1
310 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_9:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP47]], i32 4, <16 x i1> [[TMP46]], <16 x float> undef)
311 ; AVX512-NEXT: [[TMP48:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_9]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
312 ; AVX512-NEXT: [[TMP49:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 2304, i64 2320, i64 2336, i64 2352, i64 2368, i64 2384, i64 2400, i64 2416, i64 2432, i64 2448, i64 2464, i64 2480, i64 2496, i64 2512, i64 2528, i64 2544>
313 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP48]], <16 x float*> [[TMP49]], i32 4, <16 x i1> [[TMP46]])
314 ; AVX512-NEXT: [[TMP50:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 2560, i64 2576, i64 2592, i64 2608, i64 2624, i64 2640, i64 2656, i64 2672, i64 2688, i64 2704, i64 2720, i64 2736, i64 2752, i64 2768, i64 2784, i64 2800>
315 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_10:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP50]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
316 ; AVX512-NEXT: [[TMP51:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_10]], zeroinitializer
317 ; AVX512-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 2560, i64 2576, i64 2592, i64 2608, i64 2624, i64 2640, i64 2656, i64 2672, i64 2688, i64 2704, i64 2720, i64 2736, i64 2752, i64 2768, i64 2784, i64 2800>, i32 1
318 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_10:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP52]], i32 4, <16 x i1> [[TMP51]], <16 x float> undef)
319 ; AVX512-NEXT: [[TMP53:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_10]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
320 ; AVX512-NEXT: [[TMP54:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 2560, i64 2576, i64 2592, i64 2608, i64 2624, i64 2640, i64 2656, i64 2672, i64 2688, i64 2704, i64 2720, i64 2736, i64 2752, i64 2768, i64 2784, i64 2800>
321 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP53]], <16 x float*> [[TMP54]], i32 4, <16 x i1> [[TMP51]])
322 ; AVX512-NEXT: [[TMP55:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 2816, i64 2832, i64 2848, i64 2864, i64 2880, i64 2896, i64 2912, i64 2928, i64 2944, i64 2960, i64 2976, i64 2992, i64 3008, i64 3024, i64 3040, i64 3056>
323 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_11:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP55]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
324 ; AVX512-NEXT: [[TMP56:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_11]], zeroinitializer
325 ; AVX512-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 2816, i64 2832, i64 2848, i64 2864, i64 2880, i64 2896, i64 2912, i64 2928, i64 2944, i64 2960, i64 2976, i64 2992, i64 3008, i64 3024, i64 3040, i64 3056>, i32 1
326 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_11:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP57]], i32 4, <16 x i1> [[TMP56]], <16 x float> undef)
327 ; AVX512-NEXT: [[TMP58:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_11]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
328 ; AVX512-NEXT: [[TMP59:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 2816, i64 2832, i64 2848, i64 2864, i64 2880, i64 2896, i64 2912, i64 2928, i64 2944, i64 2960, i64 2976, i64 2992, i64 3008, i64 3024, i64 3040, i64 3056>
329 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP58]], <16 x float*> [[TMP59]], i32 4, <16 x i1> [[TMP56]])
330 ; AVX512-NEXT: [[TMP60:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 3072, i64 3088, i64 3104, i64 3120, i64 3136, i64 3152, i64 3168, i64 3184, i64 3200, i64 3216, i64 3232, i64 3248, i64 3264, i64 3280, i64 3296, i64 3312>
331 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_12:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP60]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
332 ; AVX512-NEXT: [[TMP61:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_12]], zeroinitializer
333 ; AVX512-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 3072, i64 3088, i64 3104, i64 3120, i64 3136, i64 3152, i64 3168, i64 3184, i64 3200, i64 3216, i64 3232, i64 3248, i64 3264, i64 3280, i64 3296, i64 3312>, i32 1
334 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_12:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP62]], i32 4, <16 x i1> [[TMP61]], <16 x float> undef)
335 ; AVX512-NEXT: [[TMP63:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_12]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
336 ; AVX512-NEXT: [[TMP64:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 3072, i64 3088, i64 3104, i64 3120, i64 3136, i64 3152, i64 3168, i64 3184, i64 3200, i64 3216, i64 3232, i64 3248, i64 3264, i64 3280, i64 3296, i64 3312>
337 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP63]], <16 x float*> [[TMP64]], i32 4, <16 x i1> [[TMP61]])
338 ; AVX512-NEXT: [[TMP65:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 3328, i64 3344, i64 3360, i64 3376, i64 3392, i64 3408, i64 3424, i64 3440, i64 3456, i64 3472, i64 3488, i64 3504, i64 3520, i64 3536, i64 3552, i64 3568>
339 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_13:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP65]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
340 ; AVX512-NEXT: [[TMP66:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_13]], zeroinitializer
341 ; AVX512-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 3328, i64 3344, i64 3360, i64 3376, i64 3392, i64 3408, i64 3424, i64 3440, i64 3456, i64 3472, i64 3488, i64 3504, i64 3520, i64 3536, i64 3552, i64 3568>, i32 1
342 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_13:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP67]], i32 4, <16 x i1> [[TMP66]], <16 x float> undef)
343 ; AVX512-NEXT: [[TMP68:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_13]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
344 ; AVX512-NEXT: [[TMP69:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 3328, i64 3344, i64 3360, i64 3376, i64 3392, i64 3408, i64 3424, i64 3440, i64 3456, i64 3472, i64 3488, i64 3504, i64 3520, i64 3536, i64 3552, i64 3568>
345 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP68]], <16 x float*> [[TMP69]], i32 4, <16 x i1> [[TMP66]])
346 ; AVX512-NEXT: [[TMP70:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 3584, i64 3600, i64 3616, i64 3632, i64 3648, i64 3664, i64 3680, i64 3696, i64 3712, i64 3728, i64 3744, i64 3760, i64 3776, i64 3792, i64 3808, i64 3824>
347 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_14:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP70]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
348 ; AVX512-NEXT: [[TMP71:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_14]], zeroinitializer
349 ; AVX512-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 3584, i64 3600, i64 3616, i64 3632, i64 3648, i64 3664, i64 3680, i64 3696, i64 3712, i64 3728, i64 3744, i64 3760, i64 3776, i64 3792, i64 3808, i64 3824>, i32 1
350 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_14:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP72]], i32 4, <16 x i1> [[TMP71]], <16 x float> undef)
351 ; AVX512-NEXT: [[TMP73:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_14]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
352 ; AVX512-NEXT: [[TMP74:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 3584, i64 3600, i64 3616, i64 3632, i64 3648, i64 3664, i64 3680, i64 3696, i64 3712, i64 3728, i64 3744, i64 3760, i64 3776, i64 3792, i64 3808, i64 3824>
353 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP73]], <16 x float*> [[TMP74]], i32 4, <16 x i1> [[TMP71]])
354 ; AVX512-NEXT: [[TMP75:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 3840, i64 3856, i64 3872, i64 3888, i64 3904, i64 3920, i64 3936, i64 3952, i64 3968, i64 3984, i64 4000, i64 4016, i64 4032, i64 4048, i64 4064, i64 4080>
355 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_15:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP75]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
356 ; AVX512-NEXT: [[TMP76:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_15]], zeroinitializer
357 ; AVX512-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 3840, i64 3856, i64 3872, i64 3888, i64 3904, i64 3920, i64 3936, i64 3952, i64 3968, i64 3984, i64 4000, i64 4016, i64 4032, i64 4048, i64 4064, i64 4080>, i32 1
358 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_15:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP77]], i32 4, <16 x i1> [[TMP76]], <16 x float> undef)
359 ; AVX512-NEXT: [[TMP78:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_15]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
360 ; AVX512-NEXT: [[TMP79:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 3840, i64 3856, i64 3872, i64 3888, i64 3904, i64 3920, i64 3936, i64 3952, i64 3968, i64 3984, i64 4000, i64 4016, i64 4032, i64 4048, i64 4064, i64 4080>
361 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP78]], <16 x float*> [[TMP79]], i32 4, <16 x i1> [[TMP76]])
362 ; AVX512-NEXT: ret void
366 ; FVW2-NEXT: br label [[VECTOR_BODY:%.*]]
368 ; FVW2-NEXT: [[INDEX6:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE30:%.*]] ]
369 ; FVW2-NEXT: [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 16>, [[ENTRY]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_STORE_CONTINUE30]] ]
370 ; FVW2-NEXT: [[STEP_ADD:%.*]] = add <2 x i64> [[VEC_IND]], <i64 32, i64 32>
371 ; FVW2-NEXT: [[STEP_ADD7:%.*]] = add <2 x i64> [[VEC_IND]], <i64 64, i64 64>
372 ; FVW2-NEXT: [[STEP_ADD8:%.*]] = add <2 x i64> [[VEC_IND]], <i64 96, i64 96>
373 ; FVW2-NEXT: [[OFFSET_IDX:%.*]] = shl i64 [[INDEX6]], 4
374 ; FVW2-NEXT: [[TMP0:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER:%.*]], <2 x i64> [[VEC_IND]]
375 ; FVW2-NEXT: [[TMP1:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <2 x i64> [[STEP_ADD]]
376 ; FVW2-NEXT: [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <2 x i64> [[STEP_ADD7]]
377 ; FVW2-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <2 x i64> [[STEP_ADD8]]
378 ; FVW2-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <2 x i32> @llvm.masked.gather.v2i32.v2p0i32(<2 x i32*> [[TMP0]], i32 4, <2 x i1> <i1 true, i1 true>, <2 x i32> undef)
379 ; FVW2-NEXT: [[WIDE_MASKED_GATHER10:%.*]] = call <2 x i32> @llvm.masked.gather.v2i32.v2p0i32(<2 x i32*> [[TMP1]], i32 4, <2 x i1> <i1 true, i1 true>, <2 x i32> undef)
380 ; FVW2-NEXT: [[WIDE_MASKED_GATHER11:%.*]] = call <2 x i32> @llvm.masked.gather.v2i32.v2p0i32(<2 x i32*> [[TMP2]], i32 4, <2 x i1> <i1 true, i1 true>, <2 x i32> undef)
381 ; FVW2-NEXT: [[WIDE_MASKED_GATHER12:%.*]] = call <2 x i32> @llvm.masked.gather.v2i32.v2p0i32(<2 x i32*> [[TMP3]], i32 4, <2 x i1> <i1 true, i1 true>, <2 x i32> undef)
382 ; FVW2-NEXT: [[TMP4:%.*]] = icmp sgt <2 x i32> [[WIDE_MASKED_GATHER]], zeroinitializer
383 ; FVW2-NEXT: [[TMP5:%.*]] = icmp sgt <2 x i32> [[WIDE_MASKED_GATHER10]], zeroinitializer
384 ; FVW2-NEXT: [[TMP6:%.*]] = icmp sgt <2 x i32> [[WIDE_MASKED_GATHER11]], zeroinitializer
385 ; FVW2-NEXT: [[TMP7:%.*]] = icmp sgt <2 x i32> [[WIDE_MASKED_GATHER12]], zeroinitializer
386 ; FVW2-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT_IN:%.*]], %struct.In* [[IN:%.*]], <2 x i64> [[VEC_IND]], i32 1
387 ; FVW2-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <2 x i64> [[STEP_ADD]], i32 1
388 ; FVW2-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <2 x i64> [[STEP_ADD7]], i32 1
389 ; FVW2-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <2 x i64> [[STEP_ADD8]], i32 1
390 ; FVW2-NEXT: [[WIDE_MASKED_GATHER13:%.*]] = call <2 x float> @llvm.masked.gather.v2f32.v2p0f32(<2 x float*> [[TMP8]], i32 4, <2 x i1> [[TMP4]], <2 x float> undef)
391 ; FVW2-NEXT: [[WIDE_MASKED_GATHER14:%.*]] = call <2 x float> @llvm.masked.gather.v2f32.v2p0f32(<2 x float*> [[TMP9]], i32 4, <2 x i1> [[TMP5]], <2 x float> undef)
392 ; FVW2-NEXT: [[WIDE_MASKED_GATHER15:%.*]] = call <2 x float> @llvm.masked.gather.v2f32.v2p0f32(<2 x float*> [[TMP10]], i32 4, <2 x i1> [[TMP6]], <2 x float> undef)
393 ; FVW2-NEXT: [[WIDE_MASKED_GATHER16:%.*]] = call <2 x float> @llvm.masked.gather.v2f32.v2p0f32(<2 x float*> [[TMP11]], i32 4, <2 x i1> [[TMP7]], <2 x float> undef)
394 ; FVW2-NEXT: [[TMP12:%.*]] = fadd <2 x float> [[WIDE_MASKED_GATHER13]], <float 5.000000e-01, float 5.000000e-01>
395 ; FVW2-NEXT: [[TMP13:%.*]] = fadd <2 x float> [[WIDE_MASKED_GATHER14]], <float 5.000000e-01, float 5.000000e-01>
396 ; FVW2-NEXT: [[TMP14:%.*]] = fadd <2 x float> [[WIDE_MASKED_GATHER15]], <float 5.000000e-01, float 5.000000e-01>
397 ; FVW2-NEXT: [[TMP15:%.*]] = fadd <2 x float> [[WIDE_MASKED_GATHER16]], <float 5.000000e-01, float 5.000000e-01>
398 ; FVW2-NEXT: [[TMP16:%.*]] = extractelement <2 x i1> [[TMP4]], i32 0
399 ; FVW2-NEXT: br i1 [[TMP16]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
400 ; FVW2: pred.store.if:
401 ; FVW2-NEXT: [[TMP17:%.*]] = getelementptr inbounds float, float* [[OUT:%.*]], i64 [[OFFSET_IDX]]
402 ; FVW2-NEXT: [[TMP18:%.*]] = extractelement <2 x float> [[TMP12]], i32 0
403 ; FVW2-NEXT: store float [[TMP18]], float* [[TMP17]], align 4
404 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE]]
405 ; FVW2: pred.store.continue:
406 ; FVW2-NEXT: [[TMP19:%.*]] = extractelement <2 x i1> [[TMP4]], i32 1
407 ; FVW2-NEXT: br i1 [[TMP19]], label [[PRED_STORE_IF17:%.*]], label [[PRED_STORE_CONTINUE18:%.*]]
408 ; FVW2: pred.store.if17:
409 ; FVW2-NEXT: [[TMP20:%.*]] = or i64 [[OFFSET_IDX]], 16
410 ; FVW2-NEXT: [[TMP21:%.*]] = getelementptr inbounds float, float* [[OUT]], i64 [[TMP20]]
411 ; FVW2-NEXT: [[TMP22:%.*]] = extractelement <2 x float> [[TMP12]], i32 1
412 ; FVW2-NEXT: store float [[TMP22]], float* [[TMP21]], align 4
413 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE18]]
414 ; FVW2: pred.store.continue18:
415 ; FVW2-NEXT: [[TMP23:%.*]] = extractelement <2 x i1> [[TMP5]], i32 0
416 ; FVW2-NEXT: br i1 [[TMP23]], label [[PRED_STORE_IF19:%.*]], label [[PRED_STORE_CONTINUE20:%.*]]
417 ; FVW2: pred.store.if19:
418 ; FVW2-NEXT: [[TMP24:%.*]] = or i64 [[OFFSET_IDX]], 32
419 ; FVW2-NEXT: [[TMP25:%.*]] = getelementptr inbounds float, float* [[OUT]], i64 [[TMP24]]
420 ; FVW2-NEXT: [[TMP26:%.*]] = extractelement <2 x float> [[TMP13]], i32 0
421 ; FVW2-NEXT: store float [[TMP26]], float* [[TMP25]], align 4
422 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE20]]
423 ; FVW2: pred.store.continue20:
424 ; FVW2-NEXT: [[TMP27:%.*]] = extractelement <2 x i1> [[TMP5]], i32 1
425 ; FVW2-NEXT: br i1 [[TMP27]], label [[PRED_STORE_IF21:%.*]], label [[PRED_STORE_CONTINUE22:%.*]]
426 ; FVW2: pred.store.if21:
427 ; FVW2-NEXT: [[TMP28:%.*]] = or i64 [[OFFSET_IDX]], 48
428 ; FVW2-NEXT: [[TMP29:%.*]] = getelementptr inbounds float, float* [[OUT]], i64 [[TMP28]]
429 ; FVW2-NEXT: [[TMP30:%.*]] = extractelement <2 x float> [[TMP13]], i32 1
430 ; FVW2-NEXT: store float [[TMP30]], float* [[TMP29]], align 4
431 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE22]]
432 ; FVW2: pred.store.continue22:
433 ; FVW2-NEXT: [[TMP31:%.*]] = extractelement <2 x i1> [[TMP6]], i32 0
434 ; FVW2-NEXT: br i1 [[TMP31]], label [[PRED_STORE_IF23:%.*]], label [[PRED_STORE_CONTINUE24:%.*]]
435 ; FVW2: pred.store.if23:
436 ; FVW2-NEXT: [[TMP32:%.*]] = or i64 [[OFFSET_IDX]], 64
437 ; FVW2-NEXT: [[TMP33:%.*]] = getelementptr inbounds float, float* [[OUT]], i64 [[TMP32]]
438 ; FVW2-NEXT: [[TMP34:%.*]] = extractelement <2 x float> [[TMP14]], i32 0
439 ; FVW2-NEXT: store float [[TMP34]], float* [[TMP33]], align 4
440 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE24]]
441 ; FVW2: pred.store.continue24:
442 ; FVW2-NEXT: [[TMP35:%.*]] = extractelement <2 x i1> [[TMP6]], i32 1
443 ; FVW2-NEXT: br i1 [[TMP35]], label [[PRED_STORE_IF25:%.*]], label [[PRED_STORE_CONTINUE26:%.*]]
444 ; FVW2: pred.store.if25:
445 ; FVW2-NEXT: [[TMP36:%.*]] = or i64 [[OFFSET_IDX]], 80
446 ; FVW2-NEXT: [[TMP37:%.*]] = getelementptr inbounds float, float* [[OUT]], i64 [[TMP36]]
447 ; FVW2-NEXT: [[TMP38:%.*]] = extractelement <2 x float> [[TMP14]], i32 1
448 ; FVW2-NEXT: store float [[TMP38]], float* [[TMP37]], align 4
449 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE26]]
450 ; FVW2: pred.store.continue26:
451 ; FVW2-NEXT: [[TMP39:%.*]] = extractelement <2 x i1> [[TMP7]], i32 0
452 ; FVW2-NEXT: br i1 [[TMP39]], label [[PRED_STORE_IF27:%.*]], label [[PRED_STORE_CONTINUE28:%.*]]
453 ; FVW2: pred.store.if27:
454 ; FVW2-NEXT: [[TMP40:%.*]] = or i64 [[OFFSET_IDX]], 96
455 ; FVW2-NEXT: [[TMP41:%.*]] = getelementptr inbounds float, float* [[OUT]], i64 [[TMP40]]
456 ; FVW2-NEXT: [[TMP42:%.*]] = extractelement <2 x float> [[TMP15]], i32 0
457 ; FVW2-NEXT: store float [[TMP42]], float* [[TMP41]], align 4
458 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE28]]
459 ; FVW2: pred.store.continue28:
460 ; FVW2-NEXT: [[TMP43:%.*]] = extractelement <2 x i1> [[TMP7]], i32 1
461 ; FVW2-NEXT: br i1 [[TMP43]], label [[PRED_STORE_IF29:%.*]], label [[PRED_STORE_CONTINUE30]]
462 ; FVW2: pred.store.if29:
463 ; FVW2-NEXT: [[TMP44:%.*]] = or i64 [[OFFSET_IDX]], 112
464 ; FVW2-NEXT: [[TMP45:%.*]] = getelementptr inbounds float, float* [[OUT]], i64 [[TMP44]]
465 ; FVW2-NEXT: [[TMP46:%.*]] = extractelement <2 x float> [[TMP15]], i32 1
466 ; FVW2-NEXT: store float [[TMP46]], float* [[TMP45]], align 4
467 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE30]]
468 ; FVW2: pred.store.continue30:
469 ; FVW2-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX6]], 8
470 ; FVW2-NEXT: [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], <i64 128, i64 128>
471 ; FVW2-NEXT: [[TMP47:%.*]] = icmp eq i64 [[INDEX_NEXT]], 256
472 ; FVW2-NEXT: br i1 [[TMP47]], label [[FOR_END:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP2:![0-9]+]]
474 ; FVW2-NEXT: ret void
477 %in.addr = alloca %struct.In*, align 8
478 %out.addr = alloca float*, align 8
479 %trigger.addr = alloca i32*, align 8
480 %index.addr = alloca i32*, align 8
481 %i = alloca i32, align 4
482 store %struct.In* %in, %struct.In** %in.addr, align 8
483 store float* %out, float** %out.addr, align 8
484 store i32* %trigger, i32** %trigger.addr, align 8
485 store i32* %index, i32** %index.addr, align 8
486 store i32 0, i32* %i, align 4
489 for.cond: ; preds = %for.inc, %entry
490 %0 = load i32, i32* %i, align 4
491 %cmp = icmp slt i32 %0, 4096
492 br i1 %cmp, label %for.body, label %for.end
494 for.body: ; preds = %for.cond
495 %1 = load i32, i32* %i, align 4
496 %idxprom = sext i32 %1 to i64
497 %2 = load i32*, i32** %trigger.addr, align 8
498 %arrayidx = getelementptr inbounds i32, i32* %2, i64 %idxprom
499 %3 = load i32, i32* %arrayidx, align 4
500 %cmp1 = icmp sgt i32 %3, 0
501 br i1 %cmp1, label %if.then, label %if.end
503 if.then: ; preds = %for.body
504 %4 = load i32, i32* %i, align 4
505 %idxprom2 = sext i32 %4 to i64
506 %5 = load %struct.In*, %struct.In** %in.addr, align 8
507 %arrayidx3 = getelementptr inbounds %struct.In, %struct.In* %5, i64 %idxprom2
508 %b = getelementptr inbounds %struct.In, %struct.In* %arrayidx3, i32 0, i32 1
509 %6 = load float, float* %b, align 4
510 %add = fadd float %6, 5.000000e-01
511 %7 = load i32, i32* %i, align 4
512 %idxprom4 = sext i32 %7 to i64
513 %8 = load float*, float** %out.addr, align 8
514 %arrayidx5 = getelementptr inbounds float, float* %8, i64 %idxprom4
515 store float %add, float* %arrayidx5, align 4
518 if.end: ; preds = %if.then, %for.body
521 for.inc: ; preds = %if.end
522 %9 = load i32, i32* %i, align 4
523 %inc = add nsw i32 %9, 16
524 store i32 %inc, i32* %i, align 4
527 for.end: ; preds = %for.cond
536 ;void foo3 (In * __restrict__ in, Out * __restrict__ out, int * __restrict__ trigger) {
538 ; for (int i=0; i<SIZE; i += 16) {
539 ; if (trigger[i] > 0) {
540 ; out[i].b = in[i].b + (float) 0.5;
545 %struct.Out = type { float, float }
547 define void @foo3(%struct.In* noalias %in, %struct.Out* noalias %out, i32* noalias %trigger) {
548 ; AVX512-LABEL: @foo3(
549 ; AVX512-NEXT: entry:
550 ; AVX512-NEXT: [[TMP0:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER:%.*]], <16 x i64> <i64 0, i64 16, i64 32, i64 48, i64 64, i64 80, i64 96, i64 112, i64 128, i64 144, i64 160, i64 176, i64 192, i64 208, i64 224, i64 240>
551 ; AVX512-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP0]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
552 ; AVX512-NEXT: [[TMP1:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER]], zeroinitializer
553 ; AVX512-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT_IN:%.*]], %struct.In* [[IN:%.*]], <16 x i64> <i64 0, i64 16, i64 32, i64 48, i64 64, i64 80, i64 96, i64 112, i64 128, i64 144, i64 160, i64 176, i64 192, i64 208, i64 224, i64 240>, i32 1
554 ; AVX512-NEXT: [[WIDE_MASKED_GATHER6:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP2]], i32 4, <16 x i1> [[TMP1]], <16 x float> undef)
555 ; AVX512-NEXT: [[TMP3:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER6]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
556 ; AVX512-NEXT: [[TMP4:%.*]] = getelementptr inbounds [[STRUCT_OUT:%.*]], %struct.Out* [[OUT:%.*]], <16 x i64> <i64 0, i64 16, i64 32, i64 48, i64 64, i64 80, i64 96, i64 112, i64 128, i64 144, i64 160, i64 176, i64 192, i64 208, i64 224, i64 240>, i32 1
557 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP3]], <16 x float*> [[TMP4]], i32 4, <16 x i1> [[TMP1]])
558 ; AVX512-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 256, i64 272, i64 288, i64 304, i64 320, i64 336, i64 352, i64 368, i64 384, i64 400, i64 416, i64 432, i64 448, i64 464, i64 480, i64 496>
559 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_1:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP5]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
560 ; AVX512-NEXT: [[TMP6:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_1]], zeroinitializer
561 ; AVX512-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 256, i64 272, i64 288, i64 304, i64 320, i64 336, i64 352, i64 368, i64 384, i64 400, i64 416, i64 432, i64 448, i64 464, i64 480, i64 496>, i32 1
562 ; AVX512-NEXT: [[WIDE_MASKED_GATHER6_1:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP7]], i32 4, <16 x i1> [[TMP6]], <16 x float> undef)
563 ; AVX512-NEXT: [[TMP8:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER6_1]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
564 ; AVX512-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT_OUT]], %struct.Out* [[OUT]], <16 x i64> <i64 256, i64 272, i64 288, i64 304, i64 320, i64 336, i64 352, i64 368, i64 384, i64 400, i64 416, i64 432, i64 448, i64 464, i64 480, i64 496>, i32 1
565 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP8]], <16 x float*> [[TMP9]], i32 4, <16 x i1> [[TMP6]])
566 ; AVX512-NEXT: [[TMP10:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 512, i64 528, i64 544, i64 560, i64 576, i64 592, i64 608, i64 624, i64 640, i64 656, i64 672, i64 688, i64 704, i64 720, i64 736, i64 752>
567 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_2:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP10]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
568 ; AVX512-NEXT: [[TMP11:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_2]], zeroinitializer
569 ; AVX512-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 512, i64 528, i64 544, i64 560, i64 576, i64 592, i64 608, i64 624, i64 640, i64 656, i64 672, i64 688, i64 704, i64 720, i64 736, i64 752>, i32 1
570 ; AVX512-NEXT: [[WIDE_MASKED_GATHER6_2:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP12]], i32 4, <16 x i1> [[TMP11]], <16 x float> undef)
571 ; AVX512-NEXT: [[TMP13:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER6_2]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
572 ; AVX512-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT_OUT]], %struct.Out* [[OUT]], <16 x i64> <i64 512, i64 528, i64 544, i64 560, i64 576, i64 592, i64 608, i64 624, i64 640, i64 656, i64 672, i64 688, i64 704, i64 720, i64 736, i64 752>, i32 1
573 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP13]], <16 x float*> [[TMP14]], i32 4, <16 x i1> [[TMP11]])
574 ; AVX512-NEXT: [[TMP15:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 768, i64 784, i64 800, i64 816, i64 832, i64 848, i64 864, i64 880, i64 896, i64 912, i64 928, i64 944, i64 960, i64 976, i64 992, i64 1008>
575 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_3:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP15]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
576 ; AVX512-NEXT: [[TMP16:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_3]], zeroinitializer
577 ; AVX512-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 768, i64 784, i64 800, i64 816, i64 832, i64 848, i64 864, i64 880, i64 896, i64 912, i64 928, i64 944, i64 960, i64 976, i64 992, i64 1008>, i32 1
578 ; AVX512-NEXT: [[WIDE_MASKED_GATHER6_3:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP17]], i32 4, <16 x i1> [[TMP16]], <16 x float> undef)
579 ; AVX512-NEXT: [[TMP18:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER6_3]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
580 ; AVX512-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[STRUCT_OUT]], %struct.Out* [[OUT]], <16 x i64> <i64 768, i64 784, i64 800, i64 816, i64 832, i64 848, i64 864, i64 880, i64 896, i64 912, i64 928, i64 944, i64 960, i64 976, i64 992, i64 1008>, i32 1
581 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP18]], <16 x float*> [[TMP19]], i32 4, <16 x i1> [[TMP16]])
582 ; AVX512-NEXT: [[TMP20:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 1024, i64 1040, i64 1056, i64 1072, i64 1088, i64 1104, i64 1120, i64 1136, i64 1152, i64 1168, i64 1184, i64 1200, i64 1216, i64 1232, i64 1248, i64 1264>
583 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_4:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP20]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
584 ; AVX512-NEXT: [[TMP21:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_4]], zeroinitializer
585 ; AVX512-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 1024, i64 1040, i64 1056, i64 1072, i64 1088, i64 1104, i64 1120, i64 1136, i64 1152, i64 1168, i64 1184, i64 1200, i64 1216, i64 1232, i64 1248, i64 1264>, i32 1
586 ; AVX512-NEXT: [[WIDE_MASKED_GATHER6_4:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP22]], i32 4, <16 x i1> [[TMP21]], <16 x float> undef)
587 ; AVX512-NEXT: [[TMP23:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER6_4]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
588 ; AVX512-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT_OUT]], %struct.Out* [[OUT]], <16 x i64> <i64 1024, i64 1040, i64 1056, i64 1072, i64 1088, i64 1104, i64 1120, i64 1136, i64 1152, i64 1168, i64 1184, i64 1200, i64 1216, i64 1232, i64 1248, i64 1264>, i32 1
589 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP23]], <16 x float*> [[TMP24]], i32 4, <16 x i1> [[TMP21]])
590 ; AVX512-NEXT: [[TMP25:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 1280, i64 1296, i64 1312, i64 1328, i64 1344, i64 1360, i64 1376, i64 1392, i64 1408, i64 1424, i64 1440, i64 1456, i64 1472, i64 1488, i64 1504, i64 1520>
591 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_5:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP25]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
592 ; AVX512-NEXT: [[TMP26:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_5]], zeroinitializer
593 ; AVX512-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 1280, i64 1296, i64 1312, i64 1328, i64 1344, i64 1360, i64 1376, i64 1392, i64 1408, i64 1424, i64 1440, i64 1456, i64 1472, i64 1488, i64 1504, i64 1520>, i32 1
594 ; AVX512-NEXT: [[WIDE_MASKED_GATHER6_5:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP27]], i32 4, <16 x i1> [[TMP26]], <16 x float> undef)
595 ; AVX512-NEXT: [[TMP28:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER6_5]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
596 ; AVX512-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT_OUT]], %struct.Out* [[OUT]], <16 x i64> <i64 1280, i64 1296, i64 1312, i64 1328, i64 1344, i64 1360, i64 1376, i64 1392, i64 1408, i64 1424, i64 1440, i64 1456, i64 1472, i64 1488, i64 1504, i64 1520>, i32 1
597 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP28]], <16 x float*> [[TMP29]], i32 4, <16 x i1> [[TMP26]])
598 ; AVX512-NEXT: [[TMP30:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 1536, i64 1552, i64 1568, i64 1584, i64 1600, i64 1616, i64 1632, i64 1648, i64 1664, i64 1680, i64 1696, i64 1712, i64 1728, i64 1744, i64 1760, i64 1776>
599 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_6:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP30]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
600 ; AVX512-NEXT: [[TMP31:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_6]], zeroinitializer
601 ; AVX512-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 1536, i64 1552, i64 1568, i64 1584, i64 1600, i64 1616, i64 1632, i64 1648, i64 1664, i64 1680, i64 1696, i64 1712, i64 1728, i64 1744, i64 1760, i64 1776>, i32 1
602 ; AVX512-NEXT: [[WIDE_MASKED_GATHER6_6:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP32]], i32 4, <16 x i1> [[TMP31]], <16 x float> undef)
603 ; AVX512-NEXT: [[TMP33:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER6_6]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
604 ; AVX512-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT_OUT]], %struct.Out* [[OUT]], <16 x i64> <i64 1536, i64 1552, i64 1568, i64 1584, i64 1600, i64 1616, i64 1632, i64 1648, i64 1664, i64 1680, i64 1696, i64 1712, i64 1728, i64 1744, i64 1760, i64 1776>, i32 1
605 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP33]], <16 x float*> [[TMP34]], i32 4, <16 x i1> [[TMP31]])
606 ; AVX512-NEXT: [[TMP35:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 1792, i64 1808, i64 1824, i64 1840, i64 1856, i64 1872, i64 1888, i64 1904, i64 1920, i64 1936, i64 1952, i64 1968, i64 1984, i64 2000, i64 2016, i64 2032>
607 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_7:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP35]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
608 ; AVX512-NEXT: [[TMP36:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_7]], zeroinitializer
609 ; AVX512-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 1792, i64 1808, i64 1824, i64 1840, i64 1856, i64 1872, i64 1888, i64 1904, i64 1920, i64 1936, i64 1952, i64 1968, i64 1984, i64 2000, i64 2016, i64 2032>, i32 1
610 ; AVX512-NEXT: [[WIDE_MASKED_GATHER6_7:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP37]], i32 4, <16 x i1> [[TMP36]], <16 x float> undef)
611 ; AVX512-NEXT: [[TMP38:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER6_7]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
612 ; AVX512-NEXT: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT_OUT]], %struct.Out* [[OUT]], <16 x i64> <i64 1792, i64 1808, i64 1824, i64 1840, i64 1856, i64 1872, i64 1888, i64 1904, i64 1920, i64 1936, i64 1952, i64 1968, i64 1984, i64 2000, i64 2016, i64 2032>, i32 1
613 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP38]], <16 x float*> [[TMP39]], i32 4, <16 x i1> [[TMP36]])
614 ; AVX512-NEXT: [[TMP40:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 2048, i64 2064, i64 2080, i64 2096, i64 2112, i64 2128, i64 2144, i64 2160, i64 2176, i64 2192, i64 2208, i64 2224, i64 2240, i64 2256, i64 2272, i64 2288>
615 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_8:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP40]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
616 ; AVX512-NEXT: [[TMP41:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_8]], zeroinitializer
617 ; AVX512-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 2048, i64 2064, i64 2080, i64 2096, i64 2112, i64 2128, i64 2144, i64 2160, i64 2176, i64 2192, i64 2208, i64 2224, i64 2240, i64 2256, i64 2272, i64 2288>, i32 1
618 ; AVX512-NEXT: [[WIDE_MASKED_GATHER6_8:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP42]], i32 4, <16 x i1> [[TMP41]], <16 x float> undef)
619 ; AVX512-NEXT: [[TMP43:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER6_8]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
620 ; AVX512-NEXT: [[TMP44:%.*]] = getelementptr inbounds [[STRUCT_OUT]], %struct.Out* [[OUT]], <16 x i64> <i64 2048, i64 2064, i64 2080, i64 2096, i64 2112, i64 2128, i64 2144, i64 2160, i64 2176, i64 2192, i64 2208, i64 2224, i64 2240, i64 2256, i64 2272, i64 2288>, i32 1
621 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP43]], <16 x float*> [[TMP44]], i32 4, <16 x i1> [[TMP41]])
622 ; AVX512-NEXT: [[TMP45:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 2304, i64 2320, i64 2336, i64 2352, i64 2368, i64 2384, i64 2400, i64 2416, i64 2432, i64 2448, i64 2464, i64 2480, i64 2496, i64 2512, i64 2528, i64 2544>
623 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_9:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP45]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
624 ; AVX512-NEXT: [[TMP46:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_9]], zeroinitializer
625 ; AVX512-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 2304, i64 2320, i64 2336, i64 2352, i64 2368, i64 2384, i64 2400, i64 2416, i64 2432, i64 2448, i64 2464, i64 2480, i64 2496, i64 2512, i64 2528, i64 2544>, i32 1
626 ; AVX512-NEXT: [[WIDE_MASKED_GATHER6_9:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP47]], i32 4, <16 x i1> [[TMP46]], <16 x float> undef)
627 ; AVX512-NEXT: [[TMP48:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER6_9]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
628 ; AVX512-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT_OUT]], %struct.Out* [[OUT]], <16 x i64> <i64 2304, i64 2320, i64 2336, i64 2352, i64 2368, i64 2384, i64 2400, i64 2416, i64 2432, i64 2448, i64 2464, i64 2480, i64 2496, i64 2512, i64 2528, i64 2544>, i32 1
629 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP48]], <16 x float*> [[TMP49]], i32 4, <16 x i1> [[TMP46]])
630 ; AVX512-NEXT: [[TMP50:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 2560, i64 2576, i64 2592, i64 2608, i64 2624, i64 2640, i64 2656, i64 2672, i64 2688, i64 2704, i64 2720, i64 2736, i64 2752, i64 2768, i64 2784, i64 2800>
631 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_10:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP50]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
632 ; AVX512-NEXT: [[TMP51:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_10]], zeroinitializer
633 ; AVX512-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 2560, i64 2576, i64 2592, i64 2608, i64 2624, i64 2640, i64 2656, i64 2672, i64 2688, i64 2704, i64 2720, i64 2736, i64 2752, i64 2768, i64 2784, i64 2800>, i32 1
634 ; AVX512-NEXT: [[WIDE_MASKED_GATHER6_10:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP52]], i32 4, <16 x i1> [[TMP51]], <16 x float> undef)
635 ; AVX512-NEXT: [[TMP53:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER6_10]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
636 ; AVX512-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT_OUT]], %struct.Out* [[OUT]], <16 x i64> <i64 2560, i64 2576, i64 2592, i64 2608, i64 2624, i64 2640, i64 2656, i64 2672, i64 2688, i64 2704, i64 2720, i64 2736, i64 2752, i64 2768, i64 2784, i64 2800>, i32 1
637 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP53]], <16 x float*> [[TMP54]], i32 4, <16 x i1> [[TMP51]])
638 ; AVX512-NEXT: [[TMP55:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 2816, i64 2832, i64 2848, i64 2864, i64 2880, i64 2896, i64 2912, i64 2928, i64 2944, i64 2960, i64 2976, i64 2992, i64 3008, i64 3024, i64 3040, i64 3056>
639 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_11:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP55]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
640 ; AVX512-NEXT: [[TMP56:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_11]], zeroinitializer
641 ; AVX512-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 2816, i64 2832, i64 2848, i64 2864, i64 2880, i64 2896, i64 2912, i64 2928, i64 2944, i64 2960, i64 2976, i64 2992, i64 3008, i64 3024, i64 3040, i64 3056>, i32 1
642 ; AVX512-NEXT: [[WIDE_MASKED_GATHER6_11:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP57]], i32 4, <16 x i1> [[TMP56]], <16 x float> undef)
643 ; AVX512-NEXT: [[TMP58:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER6_11]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
644 ; AVX512-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT_OUT]], %struct.Out* [[OUT]], <16 x i64> <i64 2816, i64 2832, i64 2848, i64 2864, i64 2880, i64 2896, i64 2912, i64 2928, i64 2944, i64 2960, i64 2976, i64 2992, i64 3008, i64 3024, i64 3040, i64 3056>, i32 1
645 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP58]], <16 x float*> [[TMP59]], i32 4, <16 x i1> [[TMP56]])
646 ; AVX512-NEXT: [[TMP60:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 3072, i64 3088, i64 3104, i64 3120, i64 3136, i64 3152, i64 3168, i64 3184, i64 3200, i64 3216, i64 3232, i64 3248, i64 3264, i64 3280, i64 3296, i64 3312>
647 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_12:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP60]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
648 ; AVX512-NEXT: [[TMP61:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_12]], zeroinitializer
649 ; AVX512-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 3072, i64 3088, i64 3104, i64 3120, i64 3136, i64 3152, i64 3168, i64 3184, i64 3200, i64 3216, i64 3232, i64 3248, i64 3264, i64 3280, i64 3296, i64 3312>, i32 1
650 ; AVX512-NEXT: [[WIDE_MASKED_GATHER6_12:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP62]], i32 4, <16 x i1> [[TMP61]], <16 x float> undef)
651 ; AVX512-NEXT: [[TMP63:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER6_12]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
652 ; AVX512-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT_OUT]], %struct.Out* [[OUT]], <16 x i64> <i64 3072, i64 3088, i64 3104, i64 3120, i64 3136, i64 3152, i64 3168, i64 3184, i64 3200, i64 3216, i64 3232, i64 3248, i64 3264, i64 3280, i64 3296, i64 3312>, i32 1
653 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP63]], <16 x float*> [[TMP64]], i32 4, <16 x i1> [[TMP61]])
654 ; AVX512-NEXT: [[TMP65:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 3328, i64 3344, i64 3360, i64 3376, i64 3392, i64 3408, i64 3424, i64 3440, i64 3456, i64 3472, i64 3488, i64 3504, i64 3520, i64 3536, i64 3552, i64 3568>
655 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_13:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP65]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
656 ; AVX512-NEXT: [[TMP66:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_13]], zeroinitializer
657 ; AVX512-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 3328, i64 3344, i64 3360, i64 3376, i64 3392, i64 3408, i64 3424, i64 3440, i64 3456, i64 3472, i64 3488, i64 3504, i64 3520, i64 3536, i64 3552, i64 3568>, i32 1
658 ; AVX512-NEXT: [[WIDE_MASKED_GATHER6_13:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP67]], i32 4, <16 x i1> [[TMP66]], <16 x float> undef)
659 ; AVX512-NEXT: [[TMP68:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER6_13]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
660 ; AVX512-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT_OUT]], %struct.Out* [[OUT]], <16 x i64> <i64 3328, i64 3344, i64 3360, i64 3376, i64 3392, i64 3408, i64 3424, i64 3440, i64 3456, i64 3472, i64 3488, i64 3504, i64 3520, i64 3536, i64 3552, i64 3568>, i32 1
661 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP68]], <16 x float*> [[TMP69]], i32 4, <16 x i1> [[TMP66]])
662 ; AVX512-NEXT: [[TMP70:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 3584, i64 3600, i64 3616, i64 3632, i64 3648, i64 3664, i64 3680, i64 3696, i64 3712, i64 3728, i64 3744, i64 3760, i64 3776, i64 3792, i64 3808, i64 3824>
663 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_14:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP70]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
664 ; AVX512-NEXT: [[TMP71:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_14]], zeroinitializer
665 ; AVX512-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 3584, i64 3600, i64 3616, i64 3632, i64 3648, i64 3664, i64 3680, i64 3696, i64 3712, i64 3728, i64 3744, i64 3760, i64 3776, i64 3792, i64 3808, i64 3824>, i32 1
666 ; AVX512-NEXT: [[WIDE_MASKED_GATHER6_14:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP72]], i32 4, <16 x i1> [[TMP71]], <16 x float> undef)
667 ; AVX512-NEXT: [[TMP73:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER6_14]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
668 ; AVX512-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT_OUT]], %struct.Out* [[OUT]], <16 x i64> <i64 3584, i64 3600, i64 3616, i64 3632, i64 3648, i64 3664, i64 3680, i64 3696, i64 3712, i64 3728, i64 3744, i64 3760, i64 3776, i64 3792, i64 3808, i64 3824>, i32 1
669 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP73]], <16 x float*> [[TMP74]], i32 4, <16 x i1> [[TMP71]])
670 ; AVX512-NEXT: [[TMP75:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 3840, i64 3856, i64 3872, i64 3888, i64 3904, i64 3920, i64 3936, i64 3952, i64 3968, i64 3984, i64 4000, i64 4016, i64 4032, i64 4048, i64 4064, i64 4080>
671 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_15:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP75]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
672 ; AVX512-NEXT: [[TMP76:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_15]], zeroinitializer
673 ; AVX512-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 3840, i64 3856, i64 3872, i64 3888, i64 3904, i64 3920, i64 3936, i64 3952, i64 3968, i64 3984, i64 4000, i64 4016, i64 4032, i64 4048, i64 4064, i64 4080>, i32 1
674 ; AVX512-NEXT: [[WIDE_MASKED_GATHER6_15:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP77]], i32 4, <16 x i1> [[TMP76]], <16 x float> undef)
675 ; AVX512-NEXT: [[TMP78:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER6_15]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
676 ; AVX512-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT_OUT]], %struct.Out* [[OUT]], <16 x i64> <i64 3840, i64 3856, i64 3872, i64 3888, i64 3904, i64 3920, i64 3936, i64 3952, i64 3968, i64 3984, i64 4000, i64 4016, i64 4032, i64 4048, i64 4064, i64 4080>, i32 1
677 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP78]], <16 x float*> [[TMP79]], i32 4, <16 x i1> [[TMP76]])
678 ; AVX512-NEXT: ret void
682 ; FVW2-NEXT: br label [[VECTOR_BODY:%.*]]
684 ; FVW2-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE29:%.*]] ]
685 ; FVW2-NEXT: [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 16>, [[ENTRY]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_STORE_CONTINUE29]] ]
686 ; FVW2-NEXT: [[STEP_ADD:%.*]] = add <2 x i64> [[VEC_IND]], <i64 32, i64 32>
687 ; FVW2-NEXT: [[STEP_ADD6:%.*]] = add <2 x i64> [[VEC_IND]], <i64 64, i64 64>
688 ; FVW2-NEXT: [[STEP_ADD7:%.*]] = add <2 x i64> [[VEC_IND]], <i64 96, i64 96>
689 ; FVW2-NEXT: [[OFFSET_IDX:%.*]] = shl i64 [[INDEX]], 4
690 ; FVW2-NEXT: [[TMP0:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER:%.*]], <2 x i64> [[VEC_IND]]
691 ; FVW2-NEXT: [[TMP1:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <2 x i64> [[STEP_ADD]]
692 ; FVW2-NEXT: [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <2 x i64> [[STEP_ADD6]]
693 ; FVW2-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <2 x i64> [[STEP_ADD7]]
694 ; FVW2-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <2 x i32> @llvm.masked.gather.v2i32.v2p0i32(<2 x i32*> [[TMP0]], i32 4, <2 x i1> <i1 true, i1 true>, <2 x i32> undef)
695 ; FVW2-NEXT: [[WIDE_MASKED_GATHER9:%.*]] = call <2 x i32> @llvm.masked.gather.v2i32.v2p0i32(<2 x i32*> [[TMP1]], i32 4, <2 x i1> <i1 true, i1 true>, <2 x i32> undef)
696 ; FVW2-NEXT: [[WIDE_MASKED_GATHER10:%.*]] = call <2 x i32> @llvm.masked.gather.v2i32.v2p0i32(<2 x i32*> [[TMP2]], i32 4, <2 x i1> <i1 true, i1 true>, <2 x i32> undef)
697 ; FVW2-NEXT: [[WIDE_MASKED_GATHER11:%.*]] = call <2 x i32> @llvm.masked.gather.v2i32.v2p0i32(<2 x i32*> [[TMP3]], i32 4, <2 x i1> <i1 true, i1 true>, <2 x i32> undef)
698 ; FVW2-NEXT: [[TMP4:%.*]] = icmp sgt <2 x i32> [[WIDE_MASKED_GATHER]], zeroinitializer
699 ; FVW2-NEXT: [[TMP5:%.*]] = icmp sgt <2 x i32> [[WIDE_MASKED_GATHER9]], zeroinitializer
700 ; FVW2-NEXT: [[TMP6:%.*]] = icmp sgt <2 x i32> [[WIDE_MASKED_GATHER10]], zeroinitializer
701 ; FVW2-NEXT: [[TMP7:%.*]] = icmp sgt <2 x i32> [[WIDE_MASKED_GATHER11]], zeroinitializer
702 ; FVW2-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT_IN:%.*]], %struct.In* [[IN:%.*]], <2 x i64> [[VEC_IND]], i32 1
703 ; FVW2-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <2 x i64> [[STEP_ADD]], i32 1
704 ; FVW2-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <2 x i64> [[STEP_ADD6]], i32 1
705 ; FVW2-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <2 x i64> [[STEP_ADD7]], i32 1
706 ; FVW2-NEXT: [[WIDE_MASKED_GATHER12:%.*]] = call <2 x float> @llvm.masked.gather.v2f32.v2p0f32(<2 x float*> [[TMP8]], i32 4, <2 x i1> [[TMP4]], <2 x float> undef)
707 ; FVW2-NEXT: [[WIDE_MASKED_GATHER13:%.*]] = call <2 x float> @llvm.masked.gather.v2f32.v2p0f32(<2 x float*> [[TMP9]], i32 4, <2 x i1> [[TMP5]], <2 x float> undef)
708 ; FVW2-NEXT: [[WIDE_MASKED_GATHER14:%.*]] = call <2 x float> @llvm.masked.gather.v2f32.v2p0f32(<2 x float*> [[TMP10]], i32 4, <2 x i1> [[TMP6]], <2 x float> undef)
709 ; FVW2-NEXT: [[WIDE_MASKED_GATHER15:%.*]] = call <2 x float> @llvm.masked.gather.v2f32.v2p0f32(<2 x float*> [[TMP11]], i32 4, <2 x i1> [[TMP7]], <2 x float> undef)
710 ; FVW2-NEXT: [[TMP12:%.*]] = fadd <2 x float> [[WIDE_MASKED_GATHER12]], <float 5.000000e-01, float 5.000000e-01>
711 ; FVW2-NEXT: [[TMP13:%.*]] = fadd <2 x float> [[WIDE_MASKED_GATHER13]], <float 5.000000e-01, float 5.000000e-01>
712 ; FVW2-NEXT: [[TMP14:%.*]] = fadd <2 x float> [[WIDE_MASKED_GATHER14]], <float 5.000000e-01, float 5.000000e-01>
713 ; FVW2-NEXT: [[TMP15:%.*]] = fadd <2 x float> [[WIDE_MASKED_GATHER15]], <float 5.000000e-01, float 5.000000e-01>
714 ; FVW2-NEXT: [[TMP16:%.*]] = extractelement <2 x i1> [[TMP4]], i32 0
715 ; FVW2-NEXT: br i1 [[TMP16]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
716 ; FVW2: pred.store.if:
717 ; FVW2-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT_OUT:%.*]], %struct.Out* [[OUT:%.*]], i64 [[OFFSET_IDX]], i32 1
718 ; FVW2-NEXT: [[TMP18:%.*]] = extractelement <2 x float> [[TMP12]], i32 0
719 ; FVW2-NEXT: store float [[TMP18]], float* [[TMP17]], align 4
720 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE]]
721 ; FVW2: pred.store.continue:
722 ; FVW2-NEXT: [[TMP19:%.*]] = extractelement <2 x i1> [[TMP4]], i32 1
723 ; FVW2-NEXT: br i1 [[TMP19]], label [[PRED_STORE_IF16:%.*]], label [[PRED_STORE_CONTINUE17:%.*]]
724 ; FVW2: pred.store.if16:
725 ; FVW2-NEXT: [[TMP20:%.*]] = or i64 [[OFFSET_IDX]], 16
726 ; FVW2-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT_OUT]], %struct.Out* [[OUT]], i64 [[TMP20]], i32 1
727 ; FVW2-NEXT: [[TMP22:%.*]] = extractelement <2 x float> [[TMP12]], i32 1
728 ; FVW2-NEXT: store float [[TMP22]], float* [[TMP21]], align 4
729 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE17]]
730 ; FVW2: pred.store.continue17:
731 ; FVW2-NEXT: [[TMP23:%.*]] = extractelement <2 x i1> [[TMP5]], i32 0
732 ; FVW2-NEXT: br i1 [[TMP23]], label [[PRED_STORE_IF18:%.*]], label [[PRED_STORE_CONTINUE19:%.*]]
733 ; FVW2: pred.store.if18:
734 ; FVW2-NEXT: [[TMP24:%.*]] = or i64 [[OFFSET_IDX]], 32
735 ; FVW2-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT_OUT]], %struct.Out* [[OUT]], i64 [[TMP24]], i32 1
736 ; FVW2-NEXT: [[TMP26:%.*]] = extractelement <2 x float> [[TMP13]], i32 0
737 ; FVW2-NEXT: store float [[TMP26]], float* [[TMP25]], align 4
738 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE19]]
739 ; FVW2: pred.store.continue19:
740 ; FVW2-NEXT: [[TMP27:%.*]] = extractelement <2 x i1> [[TMP5]], i32 1
741 ; FVW2-NEXT: br i1 [[TMP27]], label [[PRED_STORE_IF20:%.*]], label [[PRED_STORE_CONTINUE21:%.*]]
742 ; FVW2: pred.store.if20:
743 ; FVW2-NEXT: [[TMP28:%.*]] = or i64 [[OFFSET_IDX]], 48
744 ; FVW2-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT_OUT]], %struct.Out* [[OUT]], i64 [[TMP28]], i32 1
745 ; FVW2-NEXT: [[TMP30:%.*]] = extractelement <2 x float> [[TMP13]], i32 1
746 ; FVW2-NEXT: store float [[TMP30]], float* [[TMP29]], align 4
747 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE21]]
748 ; FVW2: pred.store.continue21:
749 ; FVW2-NEXT: [[TMP31:%.*]] = extractelement <2 x i1> [[TMP6]], i32 0
750 ; FVW2-NEXT: br i1 [[TMP31]], label [[PRED_STORE_IF22:%.*]], label [[PRED_STORE_CONTINUE23:%.*]]
751 ; FVW2: pred.store.if22:
752 ; FVW2-NEXT: [[TMP32:%.*]] = or i64 [[OFFSET_IDX]], 64
753 ; FVW2-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT_OUT]], %struct.Out* [[OUT]], i64 [[TMP32]], i32 1
754 ; FVW2-NEXT: [[TMP34:%.*]] = extractelement <2 x float> [[TMP14]], i32 0
755 ; FVW2-NEXT: store float [[TMP34]], float* [[TMP33]], align 4
756 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE23]]
757 ; FVW2: pred.store.continue23:
758 ; FVW2-NEXT: [[TMP35:%.*]] = extractelement <2 x i1> [[TMP6]], i32 1
759 ; FVW2-NEXT: br i1 [[TMP35]], label [[PRED_STORE_IF24:%.*]], label [[PRED_STORE_CONTINUE25:%.*]]
760 ; FVW2: pred.store.if24:
761 ; FVW2-NEXT: [[TMP36:%.*]] = or i64 [[OFFSET_IDX]], 80
762 ; FVW2-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT_OUT]], %struct.Out* [[OUT]], i64 [[TMP36]], i32 1
763 ; FVW2-NEXT: [[TMP38:%.*]] = extractelement <2 x float> [[TMP14]], i32 1
764 ; FVW2-NEXT: store float [[TMP38]], float* [[TMP37]], align 4
765 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE25]]
766 ; FVW2: pred.store.continue25:
767 ; FVW2-NEXT: [[TMP39:%.*]] = extractelement <2 x i1> [[TMP7]], i32 0
768 ; FVW2-NEXT: br i1 [[TMP39]], label [[PRED_STORE_IF26:%.*]], label [[PRED_STORE_CONTINUE27:%.*]]
769 ; FVW2: pred.store.if26:
770 ; FVW2-NEXT: [[TMP40:%.*]] = or i64 [[OFFSET_IDX]], 96
771 ; FVW2-NEXT: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT_OUT]], %struct.Out* [[OUT]], i64 [[TMP40]], i32 1
772 ; FVW2-NEXT: [[TMP42:%.*]] = extractelement <2 x float> [[TMP15]], i32 0
773 ; FVW2-NEXT: store float [[TMP42]], float* [[TMP41]], align 4
774 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE27]]
775 ; FVW2: pred.store.continue27:
776 ; FVW2-NEXT: [[TMP43:%.*]] = extractelement <2 x i1> [[TMP7]], i32 1
777 ; FVW2-NEXT: br i1 [[TMP43]], label [[PRED_STORE_IF28:%.*]], label [[PRED_STORE_CONTINUE29]]
778 ; FVW2: pred.store.if28:
779 ; FVW2-NEXT: [[TMP44:%.*]] = or i64 [[OFFSET_IDX]], 112
780 ; FVW2-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT_OUT]], %struct.Out* [[OUT]], i64 [[TMP44]], i32 1
781 ; FVW2-NEXT: [[TMP46:%.*]] = extractelement <2 x float> [[TMP15]], i32 1
782 ; FVW2-NEXT: store float [[TMP46]], float* [[TMP45]], align 4
783 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE29]]
784 ; FVW2: pred.store.continue29:
785 ; FVW2-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
786 ; FVW2-NEXT: [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], <i64 128, i64 128>
787 ; FVW2-NEXT: [[TMP47:%.*]] = icmp eq i64 [[INDEX_NEXT]], 256
788 ; FVW2-NEXT: br i1 [[TMP47]], label [[FOR_END:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
790 ; FVW2-NEXT: ret void
793 %in.addr = alloca %struct.In*, align 8
794 %out.addr = alloca %struct.Out*, align 8
795 %trigger.addr = alloca i32*, align 8
796 %i = alloca i32, align 4
797 store %struct.In* %in, %struct.In** %in.addr, align 8
798 store %struct.Out* %out, %struct.Out** %out.addr, align 8
799 store i32* %trigger, i32** %trigger.addr, align 8
800 store i32 0, i32* %i, align 4
803 for.cond: ; preds = %for.inc, %entry
804 %0 = load i32, i32* %i, align 4
805 %cmp = icmp slt i32 %0, 4096
806 br i1 %cmp, label %for.body, label %for.end
808 for.body: ; preds = %for.cond
809 %1 = load i32, i32* %i, align 4
810 %idxprom = sext i32 %1 to i64
811 %2 = load i32*, i32** %trigger.addr, align 8
812 %arrayidx = getelementptr inbounds i32, i32* %2, i64 %idxprom
813 %3 = load i32, i32* %arrayidx, align 4
814 %cmp1 = icmp sgt i32 %3, 0
815 br i1 %cmp1, label %if.then, label %if.end
817 if.then: ; preds = %for.body
818 %4 = load i32, i32* %i, align 4
819 %idxprom2 = sext i32 %4 to i64
820 %5 = load %struct.In*, %struct.In** %in.addr, align 8
821 %arrayidx3 = getelementptr inbounds %struct.In, %struct.In* %5, i64 %idxprom2
822 %b = getelementptr inbounds %struct.In, %struct.In* %arrayidx3, i32 0, i32 1
823 %6 = load float, float* %b, align 4
824 %add = fadd float %6, 5.000000e-01
825 %7 = load i32, i32* %i, align 4
826 %idxprom4 = sext i32 %7 to i64
827 %8 = load %struct.Out*, %struct.Out** %out.addr, align 8
828 %arrayidx5 = getelementptr inbounds %struct.Out, %struct.Out* %8, i64 %idxprom4
829 %b6 = getelementptr inbounds %struct.Out, %struct.Out* %arrayidx5, i32 0, i32 1
830 store float %add, float* %b6, align 4
833 if.end: ; preds = %if.then, %for.body
836 for.inc: ; preds = %if.end
837 %9 = load i32, i32* %i, align 4
838 %inc = add nsw i32 %9, 16
839 store i32 %inc, i32* %i, align 4
842 for.end: ; preds = %for.cond
845 declare void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float>, <16 x float*>, i32, <16 x i1>)
847 ; The same as @foo2 but scatter/gather argument is a vecotr of ptrs with addresspace 1
849 define void @foo2_addrspace(%struct.In addrspace(1)* noalias %in, float addrspace(1)* noalias %out, i32* noalias %trigger, i32* noalias %index) #0 {
850 ; AVX512-LABEL: @foo2_addrspace(
851 ; AVX512-NEXT: entry:
852 ; AVX512-NEXT: [[TMP0:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER:%.*]], <16 x i64> <i64 0, i64 16, i64 32, i64 48, i64 64, i64 80, i64 96, i64 112, i64 128, i64 144, i64 160, i64 176, i64 192, i64 208, i64 224, i64 240>
853 ; AVX512-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP0]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
854 ; AVX512-NEXT: [[TMP1:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER]], zeroinitializer
855 ; AVX512-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT_IN:%.*]], [[STRUCT_IN]] addrspace(1)* [[IN:%.*]], <16 x i64> <i64 0, i64 16, i64 32, i64 48, i64 64, i64 80, i64 96, i64 112, i64 128, i64 144, i64 160, i64 176, i64 192, i64 208, i64 224, i64 240>, i32 1
856 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP2]], i32 4, <16 x i1> [[TMP1]], <16 x float> undef)
857 ; AVX512-NEXT: [[TMP3:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
858 ; AVX512-NEXT: [[TMP4:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT:%.*]], <16 x i64> <i64 0, i64 16, i64 32, i64 48, i64 64, i64 80, i64 96, i64 112, i64 128, i64 144, i64 160, i64 176, i64 192, i64 208, i64 224, i64 240>
859 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP3]], <16 x float addrspace(1)*> [[TMP4]], i32 4, <16 x i1> [[TMP1]])
860 ; AVX512-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 256, i64 272, i64 288, i64 304, i64 320, i64 336, i64 352, i64 368, i64 384, i64 400, i64 416, i64 432, i64 448, i64 464, i64 480, i64 496>
861 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_1:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP5]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
862 ; AVX512-NEXT: [[TMP6:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_1]], zeroinitializer
863 ; AVX512-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 256, i64 272, i64 288, i64 304, i64 320, i64 336, i64 352, i64 368, i64 384, i64 400, i64 416, i64 432, i64 448, i64 464, i64 480, i64 496>, i32 1
864 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_1:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP7]], i32 4, <16 x i1> [[TMP6]], <16 x float> undef)
865 ; AVX512-NEXT: [[TMP8:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_1]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
866 ; AVX512-NEXT: [[TMP9:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 256, i64 272, i64 288, i64 304, i64 320, i64 336, i64 352, i64 368, i64 384, i64 400, i64 416, i64 432, i64 448, i64 464, i64 480, i64 496>
867 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP8]], <16 x float addrspace(1)*> [[TMP9]], i32 4, <16 x i1> [[TMP6]])
868 ; AVX512-NEXT: [[TMP10:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 512, i64 528, i64 544, i64 560, i64 576, i64 592, i64 608, i64 624, i64 640, i64 656, i64 672, i64 688, i64 704, i64 720, i64 736, i64 752>
869 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_2:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP10]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
870 ; AVX512-NEXT: [[TMP11:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_2]], zeroinitializer
871 ; AVX512-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 512, i64 528, i64 544, i64 560, i64 576, i64 592, i64 608, i64 624, i64 640, i64 656, i64 672, i64 688, i64 704, i64 720, i64 736, i64 752>, i32 1
872 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_2:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP12]], i32 4, <16 x i1> [[TMP11]], <16 x float> undef)
873 ; AVX512-NEXT: [[TMP13:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_2]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
874 ; AVX512-NEXT: [[TMP14:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 512, i64 528, i64 544, i64 560, i64 576, i64 592, i64 608, i64 624, i64 640, i64 656, i64 672, i64 688, i64 704, i64 720, i64 736, i64 752>
875 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP13]], <16 x float addrspace(1)*> [[TMP14]], i32 4, <16 x i1> [[TMP11]])
876 ; AVX512-NEXT: [[TMP15:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 768, i64 784, i64 800, i64 816, i64 832, i64 848, i64 864, i64 880, i64 896, i64 912, i64 928, i64 944, i64 960, i64 976, i64 992, i64 1008>
877 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_3:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP15]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
878 ; AVX512-NEXT: [[TMP16:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_3]], zeroinitializer
879 ; AVX512-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 768, i64 784, i64 800, i64 816, i64 832, i64 848, i64 864, i64 880, i64 896, i64 912, i64 928, i64 944, i64 960, i64 976, i64 992, i64 1008>, i32 1
880 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_3:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP17]], i32 4, <16 x i1> [[TMP16]], <16 x float> undef)
881 ; AVX512-NEXT: [[TMP18:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_3]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
882 ; AVX512-NEXT: [[TMP19:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 768, i64 784, i64 800, i64 816, i64 832, i64 848, i64 864, i64 880, i64 896, i64 912, i64 928, i64 944, i64 960, i64 976, i64 992, i64 1008>
883 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP18]], <16 x float addrspace(1)*> [[TMP19]], i32 4, <16 x i1> [[TMP16]])
884 ; AVX512-NEXT: [[TMP20:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 1024, i64 1040, i64 1056, i64 1072, i64 1088, i64 1104, i64 1120, i64 1136, i64 1152, i64 1168, i64 1184, i64 1200, i64 1216, i64 1232, i64 1248, i64 1264>
885 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_4:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP20]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
886 ; AVX512-NEXT: [[TMP21:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_4]], zeroinitializer
887 ; AVX512-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 1024, i64 1040, i64 1056, i64 1072, i64 1088, i64 1104, i64 1120, i64 1136, i64 1152, i64 1168, i64 1184, i64 1200, i64 1216, i64 1232, i64 1248, i64 1264>, i32 1
888 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_4:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP22]], i32 4, <16 x i1> [[TMP21]], <16 x float> undef)
889 ; AVX512-NEXT: [[TMP23:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_4]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
890 ; AVX512-NEXT: [[TMP24:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 1024, i64 1040, i64 1056, i64 1072, i64 1088, i64 1104, i64 1120, i64 1136, i64 1152, i64 1168, i64 1184, i64 1200, i64 1216, i64 1232, i64 1248, i64 1264>
891 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP23]], <16 x float addrspace(1)*> [[TMP24]], i32 4, <16 x i1> [[TMP21]])
892 ; AVX512-NEXT: [[TMP25:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 1280, i64 1296, i64 1312, i64 1328, i64 1344, i64 1360, i64 1376, i64 1392, i64 1408, i64 1424, i64 1440, i64 1456, i64 1472, i64 1488, i64 1504, i64 1520>
893 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_5:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP25]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
894 ; AVX512-NEXT: [[TMP26:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_5]], zeroinitializer
895 ; AVX512-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 1280, i64 1296, i64 1312, i64 1328, i64 1344, i64 1360, i64 1376, i64 1392, i64 1408, i64 1424, i64 1440, i64 1456, i64 1472, i64 1488, i64 1504, i64 1520>, i32 1
896 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_5:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP27]], i32 4, <16 x i1> [[TMP26]], <16 x float> undef)
897 ; AVX512-NEXT: [[TMP28:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_5]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
898 ; AVX512-NEXT: [[TMP29:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 1280, i64 1296, i64 1312, i64 1328, i64 1344, i64 1360, i64 1376, i64 1392, i64 1408, i64 1424, i64 1440, i64 1456, i64 1472, i64 1488, i64 1504, i64 1520>
899 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP28]], <16 x float addrspace(1)*> [[TMP29]], i32 4, <16 x i1> [[TMP26]])
900 ; AVX512-NEXT: [[TMP30:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 1536, i64 1552, i64 1568, i64 1584, i64 1600, i64 1616, i64 1632, i64 1648, i64 1664, i64 1680, i64 1696, i64 1712, i64 1728, i64 1744, i64 1760, i64 1776>
901 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_6:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP30]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
902 ; AVX512-NEXT: [[TMP31:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_6]], zeroinitializer
903 ; AVX512-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 1536, i64 1552, i64 1568, i64 1584, i64 1600, i64 1616, i64 1632, i64 1648, i64 1664, i64 1680, i64 1696, i64 1712, i64 1728, i64 1744, i64 1760, i64 1776>, i32 1
904 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_6:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP32]], i32 4, <16 x i1> [[TMP31]], <16 x float> undef)
905 ; AVX512-NEXT: [[TMP33:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_6]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
906 ; AVX512-NEXT: [[TMP34:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 1536, i64 1552, i64 1568, i64 1584, i64 1600, i64 1616, i64 1632, i64 1648, i64 1664, i64 1680, i64 1696, i64 1712, i64 1728, i64 1744, i64 1760, i64 1776>
907 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP33]], <16 x float addrspace(1)*> [[TMP34]], i32 4, <16 x i1> [[TMP31]])
908 ; AVX512-NEXT: [[TMP35:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 1792, i64 1808, i64 1824, i64 1840, i64 1856, i64 1872, i64 1888, i64 1904, i64 1920, i64 1936, i64 1952, i64 1968, i64 1984, i64 2000, i64 2016, i64 2032>
909 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_7:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP35]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
910 ; AVX512-NEXT: [[TMP36:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_7]], zeroinitializer
911 ; AVX512-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 1792, i64 1808, i64 1824, i64 1840, i64 1856, i64 1872, i64 1888, i64 1904, i64 1920, i64 1936, i64 1952, i64 1968, i64 1984, i64 2000, i64 2016, i64 2032>, i32 1
912 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_7:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP37]], i32 4, <16 x i1> [[TMP36]], <16 x float> undef)
913 ; AVX512-NEXT: [[TMP38:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_7]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
914 ; AVX512-NEXT: [[TMP39:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 1792, i64 1808, i64 1824, i64 1840, i64 1856, i64 1872, i64 1888, i64 1904, i64 1920, i64 1936, i64 1952, i64 1968, i64 1984, i64 2000, i64 2016, i64 2032>
915 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP38]], <16 x float addrspace(1)*> [[TMP39]], i32 4, <16 x i1> [[TMP36]])
916 ; AVX512-NEXT: [[TMP40:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 2048, i64 2064, i64 2080, i64 2096, i64 2112, i64 2128, i64 2144, i64 2160, i64 2176, i64 2192, i64 2208, i64 2224, i64 2240, i64 2256, i64 2272, i64 2288>
917 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_8:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP40]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
918 ; AVX512-NEXT: [[TMP41:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_8]], zeroinitializer
919 ; AVX512-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 2048, i64 2064, i64 2080, i64 2096, i64 2112, i64 2128, i64 2144, i64 2160, i64 2176, i64 2192, i64 2208, i64 2224, i64 2240, i64 2256, i64 2272, i64 2288>, i32 1
920 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_8:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP42]], i32 4, <16 x i1> [[TMP41]], <16 x float> undef)
921 ; AVX512-NEXT: [[TMP43:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_8]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
922 ; AVX512-NEXT: [[TMP44:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 2048, i64 2064, i64 2080, i64 2096, i64 2112, i64 2128, i64 2144, i64 2160, i64 2176, i64 2192, i64 2208, i64 2224, i64 2240, i64 2256, i64 2272, i64 2288>
923 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP43]], <16 x float addrspace(1)*> [[TMP44]], i32 4, <16 x i1> [[TMP41]])
924 ; AVX512-NEXT: [[TMP45:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 2304, i64 2320, i64 2336, i64 2352, i64 2368, i64 2384, i64 2400, i64 2416, i64 2432, i64 2448, i64 2464, i64 2480, i64 2496, i64 2512, i64 2528, i64 2544>
925 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_9:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP45]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
926 ; AVX512-NEXT: [[TMP46:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_9]], zeroinitializer
927 ; AVX512-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 2304, i64 2320, i64 2336, i64 2352, i64 2368, i64 2384, i64 2400, i64 2416, i64 2432, i64 2448, i64 2464, i64 2480, i64 2496, i64 2512, i64 2528, i64 2544>, i32 1
928 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_9:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP47]], i32 4, <16 x i1> [[TMP46]], <16 x float> undef)
929 ; AVX512-NEXT: [[TMP48:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_9]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
930 ; AVX512-NEXT: [[TMP49:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 2304, i64 2320, i64 2336, i64 2352, i64 2368, i64 2384, i64 2400, i64 2416, i64 2432, i64 2448, i64 2464, i64 2480, i64 2496, i64 2512, i64 2528, i64 2544>
931 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP48]], <16 x float addrspace(1)*> [[TMP49]], i32 4, <16 x i1> [[TMP46]])
932 ; AVX512-NEXT: [[TMP50:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 2560, i64 2576, i64 2592, i64 2608, i64 2624, i64 2640, i64 2656, i64 2672, i64 2688, i64 2704, i64 2720, i64 2736, i64 2752, i64 2768, i64 2784, i64 2800>
933 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_10:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP50]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
934 ; AVX512-NEXT: [[TMP51:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_10]], zeroinitializer
935 ; AVX512-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 2560, i64 2576, i64 2592, i64 2608, i64 2624, i64 2640, i64 2656, i64 2672, i64 2688, i64 2704, i64 2720, i64 2736, i64 2752, i64 2768, i64 2784, i64 2800>, i32 1
936 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_10:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP52]], i32 4, <16 x i1> [[TMP51]], <16 x float> undef)
937 ; AVX512-NEXT: [[TMP53:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_10]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
938 ; AVX512-NEXT: [[TMP54:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 2560, i64 2576, i64 2592, i64 2608, i64 2624, i64 2640, i64 2656, i64 2672, i64 2688, i64 2704, i64 2720, i64 2736, i64 2752, i64 2768, i64 2784, i64 2800>
939 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP53]], <16 x float addrspace(1)*> [[TMP54]], i32 4, <16 x i1> [[TMP51]])
940 ; AVX512-NEXT: [[TMP55:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 2816, i64 2832, i64 2848, i64 2864, i64 2880, i64 2896, i64 2912, i64 2928, i64 2944, i64 2960, i64 2976, i64 2992, i64 3008, i64 3024, i64 3040, i64 3056>
941 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_11:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP55]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
942 ; AVX512-NEXT: [[TMP56:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_11]], zeroinitializer
943 ; AVX512-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 2816, i64 2832, i64 2848, i64 2864, i64 2880, i64 2896, i64 2912, i64 2928, i64 2944, i64 2960, i64 2976, i64 2992, i64 3008, i64 3024, i64 3040, i64 3056>, i32 1
944 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_11:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP57]], i32 4, <16 x i1> [[TMP56]], <16 x float> undef)
945 ; AVX512-NEXT: [[TMP58:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_11]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
946 ; AVX512-NEXT: [[TMP59:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 2816, i64 2832, i64 2848, i64 2864, i64 2880, i64 2896, i64 2912, i64 2928, i64 2944, i64 2960, i64 2976, i64 2992, i64 3008, i64 3024, i64 3040, i64 3056>
947 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP58]], <16 x float addrspace(1)*> [[TMP59]], i32 4, <16 x i1> [[TMP56]])
948 ; AVX512-NEXT: [[TMP60:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 3072, i64 3088, i64 3104, i64 3120, i64 3136, i64 3152, i64 3168, i64 3184, i64 3200, i64 3216, i64 3232, i64 3248, i64 3264, i64 3280, i64 3296, i64 3312>
949 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_12:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP60]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
950 ; AVX512-NEXT: [[TMP61:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_12]], zeroinitializer
951 ; AVX512-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 3072, i64 3088, i64 3104, i64 3120, i64 3136, i64 3152, i64 3168, i64 3184, i64 3200, i64 3216, i64 3232, i64 3248, i64 3264, i64 3280, i64 3296, i64 3312>, i32 1
952 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_12:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP62]], i32 4, <16 x i1> [[TMP61]], <16 x float> undef)
953 ; AVX512-NEXT: [[TMP63:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_12]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
954 ; AVX512-NEXT: [[TMP64:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 3072, i64 3088, i64 3104, i64 3120, i64 3136, i64 3152, i64 3168, i64 3184, i64 3200, i64 3216, i64 3232, i64 3248, i64 3264, i64 3280, i64 3296, i64 3312>
955 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP63]], <16 x float addrspace(1)*> [[TMP64]], i32 4, <16 x i1> [[TMP61]])
956 ; AVX512-NEXT: [[TMP65:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 3328, i64 3344, i64 3360, i64 3376, i64 3392, i64 3408, i64 3424, i64 3440, i64 3456, i64 3472, i64 3488, i64 3504, i64 3520, i64 3536, i64 3552, i64 3568>
957 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_13:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP65]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
958 ; AVX512-NEXT: [[TMP66:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_13]], zeroinitializer
959 ; AVX512-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 3328, i64 3344, i64 3360, i64 3376, i64 3392, i64 3408, i64 3424, i64 3440, i64 3456, i64 3472, i64 3488, i64 3504, i64 3520, i64 3536, i64 3552, i64 3568>, i32 1
960 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_13:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP67]], i32 4, <16 x i1> [[TMP66]], <16 x float> undef)
961 ; AVX512-NEXT: [[TMP68:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_13]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
962 ; AVX512-NEXT: [[TMP69:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 3328, i64 3344, i64 3360, i64 3376, i64 3392, i64 3408, i64 3424, i64 3440, i64 3456, i64 3472, i64 3488, i64 3504, i64 3520, i64 3536, i64 3552, i64 3568>
963 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP68]], <16 x float addrspace(1)*> [[TMP69]], i32 4, <16 x i1> [[TMP66]])
964 ; AVX512-NEXT: [[TMP70:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 3584, i64 3600, i64 3616, i64 3632, i64 3648, i64 3664, i64 3680, i64 3696, i64 3712, i64 3728, i64 3744, i64 3760, i64 3776, i64 3792, i64 3808, i64 3824>
965 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_14:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP70]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
966 ; AVX512-NEXT: [[TMP71:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_14]], zeroinitializer
967 ; AVX512-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 3584, i64 3600, i64 3616, i64 3632, i64 3648, i64 3664, i64 3680, i64 3696, i64 3712, i64 3728, i64 3744, i64 3760, i64 3776, i64 3792, i64 3808, i64 3824>, i32 1
968 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_14:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP72]], i32 4, <16 x i1> [[TMP71]], <16 x float> undef)
969 ; AVX512-NEXT: [[TMP73:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_14]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
970 ; AVX512-NEXT: [[TMP74:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 3584, i64 3600, i64 3616, i64 3632, i64 3648, i64 3664, i64 3680, i64 3696, i64 3712, i64 3728, i64 3744, i64 3760, i64 3776, i64 3792, i64 3808, i64 3824>
971 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP73]], <16 x float addrspace(1)*> [[TMP74]], i32 4, <16 x i1> [[TMP71]])
972 ; AVX512-NEXT: [[TMP75:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 3840, i64 3856, i64 3872, i64 3888, i64 3904, i64 3920, i64 3936, i64 3952, i64 3968, i64 3984, i64 4000, i64 4016, i64 4032, i64 4048, i64 4064, i64 4080>
973 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_15:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP75]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
974 ; AVX512-NEXT: [[TMP76:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_15]], zeroinitializer
975 ; AVX512-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 3840, i64 3856, i64 3872, i64 3888, i64 3904, i64 3920, i64 3936, i64 3952, i64 3968, i64 3984, i64 4000, i64 4016, i64 4032, i64 4048, i64 4064, i64 4080>, i32 1
976 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_15:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP77]], i32 4, <16 x i1> [[TMP76]], <16 x float> undef)
977 ; AVX512-NEXT: [[TMP78:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_15]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
978 ; AVX512-NEXT: [[TMP79:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 3840, i64 3856, i64 3872, i64 3888, i64 3904, i64 3920, i64 3936, i64 3952, i64 3968, i64 3984, i64 4000, i64 4016, i64 4032, i64 4048, i64 4064, i64 4080>
979 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP78]], <16 x float addrspace(1)*> [[TMP79]], i32 4, <16 x i1> [[TMP76]])
980 ; AVX512-NEXT: ret void
982 ; FVW2-LABEL: @foo2_addrspace(
984 ; FVW2-NEXT: br label [[VECTOR_BODY:%.*]]
986 ; FVW2-NEXT: [[INDEX6:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE30:%.*]] ]
987 ; FVW2-NEXT: [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 16>, [[ENTRY]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_STORE_CONTINUE30]] ]
988 ; FVW2-NEXT: [[STEP_ADD:%.*]] = add <2 x i64> [[VEC_IND]], <i64 32, i64 32>
989 ; FVW2-NEXT: [[STEP_ADD7:%.*]] = add <2 x i64> [[VEC_IND]], <i64 64, i64 64>
990 ; FVW2-NEXT: [[STEP_ADD8:%.*]] = add <2 x i64> [[VEC_IND]], <i64 96, i64 96>
991 ; FVW2-NEXT: [[OFFSET_IDX:%.*]] = shl i64 [[INDEX6]], 4
992 ; FVW2-NEXT: [[TMP0:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER:%.*]], <2 x i64> [[VEC_IND]]
993 ; FVW2-NEXT: [[TMP1:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <2 x i64> [[STEP_ADD]]
994 ; FVW2-NEXT: [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <2 x i64> [[STEP_ADD7]]
995 ; FVW2-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <2 x i64> [[STEP_ADD8]]
996 ; FVW2-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <2 x i32> @llvm.masked.gather.v2i32.v2p0i32(<2 x i32*> [[TMP0]], i32 4, <2 x i1> <i1 true, i1 true>, <2 x i32> undef)
997 ; FVW2-NEXT: [[WIDE_MASKED_GATHER10:%.*]] = call <2 x i32> @llvm.masked.gather.v2i32.v2p0i32(<2 x i32*> [[TMP1]], i32 4, <2 x i1> <i1 true, i1 true>, <2 x i32> undef)
998 ; FVW2-NEXT: [[WIDE_MASKED_GATHER11:%.*]] = call <2 x i32> @llvm.masked.gather.v2i32.v2p0i32(<2 x i32*> [[TMP2]], i32 4, <2 x i1> <i1 true, i1 true>, <2 x i32> undef)
999 ; FVW2-NEXT: [[WIDE_MASKED_GATHER12:%.*]] = call <2 x i32> @llvm.masked.gather.v2i32.v2p0i32(<2 x i32*> [[TMP3]], i32 4, <2 x i1> <i1 true, i1 true>, <2 x i32> undef)
1000 ; FVW2-NEXT: [[TMP4:%.*]] = icmp sgt <2 x i32> [[WIDE_MASKED_GATHER]], zeroinitializer
1001 ; FVW2-NEXT: [[TMP5:%.*]] = icmp sgt <2 x i32> [[WIDE_MASKED_GATHER10]], zeroinitializer
1002 ; FVW2-NEXT: [[TMP6:%.*]] = icmp sgt <2 x i32> [[WIDE_MASKED_GATHER11]], zeroinitializer
1003 ; FVW2-NEXT: [[TMP7:%.*]] = icmp sgt <2 x i32> [[WIDE_MASKED_GATHER12]], zeroinitializer
1004 ; FVW2-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT_IN:%.*]], [[STRUCT_IN]] addrspace(1)* [[IN:%.*]], <2 x i64> [[VEC_IND]], i32 1
1005 ; FVW2-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <2 x i64> [[STEP_ADD]], i32 1
1006 ; FVW2-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <2 x i64> [[STEP_ADD7]], i32 1
1007 ; FVW2-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <2 x i64> [[STEP_ADD8]], i32 1
1008 ; FVW2-NEXT: [[WIDE_MASKED_GATHER13:%.*]] = call <2 x float> @llvm.masked.gather.v2f32.v2p1f32(<2 x float addrspace(1)*> [[TMP8]], i32 4, <2 x i1> [[TMP4]], <2 x float> undef)
1009 ; FVW2-NEXT: [[WIDE_MASKED_GATHER14:%.*]] = call <2 x float> @llvm.masked.gather.v2f32.v2p1f32(<2 x float addrspace(1)*> [[TMP9]], i32 4, <2 x i1> [[TMP5]], <2 x float> undef)
1010 ; FVW2-NEXT: [[WIDE_MASKED_GATHER15:%.*]] = call <2 x float> @llvm.masked.gather.v2f32.v2p1f32(<2 x float addrspace(1)*> [[TMP10]], i32 4, <2 x i1> [[TMP6]], <2 x float> undef)
1011 ; FVW2-NEXT: [[WIDE_MASKED_GATHER16:%.*]] = call <2 x float> @llvm.masked.gather.v2f32.v2p1f32(<2 x float addrspace(1)*> [[TMP11]], i32 4, <2 x i1> [[TMP7]], <2 x float> undef)
1012 ; FVW2-NEXT: [[TMP12:%.*]] = fadd <2 x float> [[WIDE_MASKED_GATHER13]], <float 5.000000e-01, float 5.000000e-01>
1013 ; FVW2-NEXT: [[TMP13:%.*]] = fadd <2 x float> [[WIDE_MASKED_GATHER14]], <float 5.000000e-01, float 5.000000e-01>
1014 ; FVW2-NEXT: [[TMP14:%.*]] = fadd <2 x float> [[WIDE_MASKED_GATHER15]], <float 5.000000e-01, float 5.000000e-01>
1015 ; FVW2-NEXT: [[TMP15:%.*]] = fadd <2 x float> [[WIDE_MASKED_GATHER16]], <float 5.000000e-01, float 5.000000e-01>
1016 ; FVW2-NEXT: [[TMP16:%.*]] = extractelement <2 x i1> [[TMP4]], i32 0
1017 ; FVW2-NEXT: br i1 [[TMP16]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
1018 ; FVW2: pred.store.if:
1019 ; FVW2-NEXT: [[TMP17:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT:%.*]], i64 [[OFFSET_IDX]]
1020 ; FVW2-NEXT: [[TMP18:%.*]] = extractelement <2 x float> [[TMP12]], i32 0
1021 ; FVW2-NEXT: store float [[TMP18]], float addrspace(1)* [[TMP17]], align 4
1022 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE]]
1023 ; FVW2: pred.store.continue:
1024 ; FVW2-NEXT: [[TMP19:%.*]] = extractelement <2 x i1> [[TMP4]], i32 1
1025 ; FVW2-NEXT: br i1 [[TMP19]], label [[PRED_STORE_IF17:%.*]], label [[PRED_STORE_CONTINUE18:%.*]]
1026 ; FVW2: pred.store.if17:
1027 ; FVW2-NEXT: [[TMP20:%.*]] = or i64 [[OFFSET_IDX]], 16
1028 ; FVW2-NEXT: [[TMP21:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], i64 [[TMP20]]
1029 ; FVW2-NEXT: [[TMP22:%.*]] = extractelement <2 x float> [[TMP12]], i32 1
1030 ; FVW2-NEXT: store float [[TMP22]], float addrspace(1)* [[TMP21]], align 4
1031 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE18]]
1032 ; FVW2: pred.store.continue18:
1033 ; FVW2-NEXT: [[TMP23:%.*]] = extractelement <2 x i1> [[TMP5]], i32 0
1034 ; FVW2-NEXT: br i1 [[TMP23]], label [[PRED_STORE_IF19:%.*]], label [[PRED_STORE_CONTINUE20:%.*]]
1035 ; FVW2: pred.store.if19:
1036 ; FVW2-NEXT: [[TMP24:%.*]] = or i64 [[OFFSET_IDX]], 32
1037 ; FVW2-NEXT: [[TMP25:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], i64 [[TMP24]]
1038 ; FVW2-NEXT: [[TMP26:%.*]] = extractelement <2 x float> [[TMP13]], i32 0
1039 ; FVW2-NEXT: store float [[TMP26]], float addrspace(1)* [[TMP25]], align 4
1040 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE20]]
1041 ; FVW2: pred.store.continue20:
1042 ; FVW2-NEXT: [[TMP27:%.*]] = extractelement <2 x i1> [[TMP5]], i32 1
1043 ; FVW2-NEXT: br i1 [[TMP27]], label [[PRED_STORE_IF21:%.*]], label [[PRED_STORE_CONTINUE22:%.*]]
1044 ; FVW2: pred.store.if21:
1045 ; FVW2-NEXT: [[TMP28:%.*]] = or i64 [[OFFSET_IDX]], 48
1046 ; FVW2-NEXT: [[TMP29:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], i64 [[TMP28]]
1047 ; FVW2-NEXT: [[TMP30:%.*]] = extractelement <2 x float> [[TMP13]], i32 1
1048 ; FVW2-NEXT: store float [[TMP30]], float addrspace(1)* [[TMP29]], align 4
1049 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE22]]
1050 ; FVW2: pred.store.continue22:
1051 ; FVW2-NEXT: [[TMP31:%.*]] = extractelement <2 x i1> [[TMP6]], i32 0
1052 ; FVW2-NEXT: br i1 [[TMP31]], label [[PRED_STORE_IF23:%.*]], label [[PRED_STORE_CONTINUE24:%.*]]
1053 ; FVW2: pred.store.if23:
1054 ; FVW2-NEXT: [[TMP32:%.*]] = or i64 [[OFFSET_IDX]], 64
1055 ; FVW2-NEXT: [[TMP33:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], i64 [[TMP32]]
1056 ; FVW2-NEXT: [[TMP34:%.*]] = extractelement <2 x float> [[TMP14]], i32 0
1057 ; FVW2-NEXT: store float [[TMP34]], float addrspace(1)* [[TMP33]], align 4
1058 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE24]]
1059 ; FVW2: pred.store.continue24:
1060 ; FVW2-NEXT: [[TMP35:%.*]] = extractelement <2 x i1> [[TMP6]], i32 1
1061 ; FVW2-NEXT: br i1 [[TMP35]], label [[PRED_STORE_IF25:%.*]], label [[PRED_STORE_CONTINUE26:%.*]]
1062 ; FVW2: pred.store.if25:
1063 ; FVW2-NEXT: [[TMP36:%.*]] = or i64 [[OFFSET_IDX]], 80
1064 ; FVW2-NEXT: [[TMP37:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], i64 [[TMP36]]
1065 ; FVW2-NEXT: [[TMP38:%.*]] = extractelement <2 x float> [[TMP14]], i32 1
1066 ; FVW2-NEXT: store float [[TMP38]], float addrspace(1)* [[TMP37]], align 4
1067 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE26]]
1068 ; FVW2: pred.store.continue26:
1069 ; FVW2-NEXT: [[TMP39:%.*]] = extractelement <2 x i1> [[TMP7]], i32 0
1070 ; FVW2-NEXT: br i1 [[TMP39]], label [[PRED_STORE_IF27:%.*]], label [[PRED_STORE_CONTINUE28:%.*]]
1071 ; FVW2: pred.store.if27:
1072 ; FVW2-NEXT: [[TMP40:%.*]] = or i64 [[OFFSET_IDX]], 96
1073 ; FVW2-NEXT: [[TMP41:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], i64 [[TMP40]]
1074 ; FVW2-NEXT: [[TMP42:%.*]] = extractelement <2 x float> [[TMP15]], i32 0
1075 ; FVW2-NEXT: store float [[TMP42]], float addrspace(1)* [[TMP41]], align 4
1076 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE28]]
1077 ; FVW2: pred.store.continue28:
1078 ; FVW2-NEXT: [[TMP43:%.*]] = extractelement <2 x i1> [[TMP7]], i32 1
1079 ; FVW2-NEXT: br i1 [[TMP43]], label [[PRED_STORE_IF29:%.*]], label [[PRED_STORE_CONTINUE30]]
1080 ; FVW2: pred.store.if29:
1081 ; FVW2-NEXT: [[TMP44:%.*]] = or i64 [[OFFSET_IDX]], 112
1082 ; FVW2-NEXT: [[TMP45:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], i64 [[TMP44]]
1083 ; FVW2-NEXT: [[TMP46:%.*]] = extractelement <2 x float> [[TMP15]], i32 1
1084 ; FVW2-NEXT: store float [[TMP46]], float addrspace(1)* [[TMP45]], align 4
1085 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE30]]
1086 ; FVW2: pred.store.continue30:
1087 ; FVW2-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX6]], 8
1088 ; FVW2-NEXT: [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], <i64 128, i64 128>
1089 ; FVW2-NEXT: [[TMP47:%.*]] = icmp eq i64 [[INDEX_NEXT]], 256
1090 ; FVW2-NEXT: br i1 [[TMP47]], label [[FOR_END:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
1092 ; FVW2-NEXT: ret void
1095 %in.addr = alloca %struct.In addrspace(1)*, align 8
1096 %out.addr = alloca float addrspace(1)*, align 8
1097 %trigger.addr = alloca i32*, align 8
1098 %index.addr = alloca i32*, align 8
1099 %i = alloca i32, align 4
1100 store %struct.In addrspace(1)* %in, %struct.In addrspace(1)** %in.addr, align 8
1101 store float addrspace(1)* %out, float addrspace(1)** %out.addr, align 8
1102 store i32* %trigger, i32** %trigger.addr, align 8
1103 store i32* %index, i32** %index.addr, align 8
1104 store i32 0, i32* %i, align 4
1107 for.cond: ; preds = %for.inc, %entry
1108 %0 = load i32, i32* %i, align 4
1109 %cmp = icmp slt i32 %0, 4096
1110 br i1 %cmp, label %for.body, label %for.end
1112 for.body: ; preds = %for.cond
1113 %1 = load i32, i32* %i, align 4
1114 %idxprom = sext i32 %1 to i64
1115 %2 = load i32*, i32** %trigger.addr, align 8
1116 %arrayidx = getelementptr inbounds i32, i32* %2, i64 %idxprom
1117 %3 = load i32, i32* %arrayidx, align 4
1118 %cmp1 = icmp sgt i32 %3, 0
1119 br i1 %cmp1, label %if.then, label %if.end
1121 if.then: ; preds = %for.body
1122 %4 = load i32, i32* %i, align 4
1123 %idxprom2 = sext i32 %4 to i64
1124 %5 = load %struct.In addrspace(1)*, %struct.In addrspace(1)** %in.addr, align 8
1125 %arrayidx3 = getelementptr inbounds %struct.In, %struct.In addrspace(1)* %5, i64 %idxprom2
1126 %b = getelementptr inbounds %struct.In, %struct.In addrspace(1)* %arrayidx3, i32 0, i32 1
1127 %6 = load float, float addrspace(1)* %b, align 4
1128 %add = fadd float %6, 5.000000e-01
1129 %7 = load i32, i32* %i, align 4
1130 %idxprom4 = sext i32 %7 to i64
1131 %8 = load float addrspace(1)*, float addrspace(1)** %out.addr, align 8
1132 %arrayidx5 = getelementptr inbounds float, float addrspace(1)* %8, i64 %idxprom4
1133 store float %add, float addrspace(1)* %arrayidx5, align 4
1136 if.end: ; preds = %if.then, %for.body
1139 for.inc: ; preds = %if.end
1140 %9 = load i32, i32* %i, align 4
1141 %inc = add nsw i32 %9, 16
1142 store i32 %inc, i32* %i, align 4
1145 for.end: ; preds = %for.cond
1149 ; Same as foo2_addrspace but here only the input has the non-default address space.
1151 define void @foo2_addrspace2(%struct.In addrspace(1)* noalias %in, float addrspace(0)* noalias %out, i32* noalias %trigger, i32* noalias %index) {
1152 ; AVX512-LABEL: @foo2_addrspace2(
1153 ; AVX512-NEXT: entry:
1154 ; AVX512-NEXT: [[TMP0:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER:%.*]], <16 x i64> <i64 0, i64 16, i64 32, i64 48, i64 64, i64 80, i64 96, i64 112, i64 128, i64 144, i64 160, i64 176, i64 192, i64 208, i64 224, i64 240>
1155 ; AVX512-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP0]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1156 ; AVX512-NEXT: [[TMP1:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER]], zeroinitializer
1157 ; AVX512-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT_IN:%.*]], [[STRUCT_IN]] addrspace(1)* [[IN:%.*]], <16 x i64> <i64 0, i64 16, i64 32, i64 48, i64 64, i64 80, i64 96, i64 112, i64 128, i64 144, i64 160, i64 176, i64 192, i64 208, i64 224, i64 240>, i32 1
1158 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP2]], i32 4, <16 x i1> [[TMP1]], <16 x float> undef)
1159 ; AVX512-NEXT: [[TMP3:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1160 ; AVX512-NEXT: [[TMP4:%.*]] = getelementptr inbounds float, float* [[OUT:%.*]], <16 x i64> <i64 0, i64 16, i64 32, i64 48, i64 64, i64 80, i64 96, i64 112, i64 128, i64 144, i64 160, i64 176, i64 192, i64 208, i64 224, i64 240>
1161 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP3]], <16 x float*> [[TMP4]], i32 4, <16 x i1> [[TMP1]])
1162 ; AVX512-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 256, i64 272, i64 288, i64 304, i64 320, i64 336, i64 352, i64 368, i64 384, i64 400, i64 416, i64 432, i64 448, i64 464, i64 480, i64 496>
1163 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_1:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP5]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1164 ; AVX512-NEXT: [[TMP6:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_1]], zeroinitializer
1165 ; AVX512-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 256, i64 272, i64 288, i64 304, i64 320, i64 336, i64 352, i64 368, i64 384, i64 400, i64 416, i64 432, i64 448, i64 464, i64 480, i64 496>, i32 1
1166 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_1:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP7]], i32 4, <16 x i1> [[TMP6]], <16 x float> undef)
1167 ; AVX512-NEXT: [[TMP8:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_1]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1168 ; AVX512-NEXT: [[TMP9:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 256, i64 272, i64 288, i64 304, i64 320, i64 336, i64 352, i64 368, i64 384, i64 400, i64 416, i64 432, i64 448, i64 464, i64 480, i64 496>
1169 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP8]], <16 x float*> [[TMP9]], i32 4, <16 x i1> [[TMP6]])
1170 ; AVX512-NEXT: [[TMP10:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 512, i64 528, i64 544, i64 560, i64 576, i64 592, i64 608, i64 624, i64 640, i64 656, i64 672, i64 688, i64 704, i64 720, i64 736, i64 752>
1171 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_2:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP10]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1172 ; AVX512-NEXT: [[TMP11:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_2]], zeroinitializer
1173 ; AVX512-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 512, i64 528, i64 544, i64 560, i64 576, i64 592, i64 608, i64 624, i64 640, i64 656, i64 672, i64 688, i64 704, i64 720, i64 736, i64 752>, i32 1
1174 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_2:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP12]], i32 4, <16 x i1> [[TMP11]], <16 x float> undef)
1175 ; AVX512-NEXT: [[TMP13:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_2]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1176 ; AVX512-NEXT: [[TMP14:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 512, i64 528, i64 544, i64 560, i64 576, i64 592, i64 608, i64 624, i64 640, i64 656, i64 672, i64 688, i64 704, i64 720, i64 736, i64 752>
1177 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP13]], <16 x float*> [[TMP14]], i32 4, <16 x i1> [[TMP11]])
1178 ; AVX512-NEXT: [[TMP15:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 768, i64 784, i64 800, i64 816, i64 832, i64 848, i64 864, i64 880, i64 896, i64 912, i64 928, i64 944, i64 960, i64 976, i64 992, i64 1008>
1179 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_3:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP15]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1180 ; AVX512-NEXT: [[TMP16:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_3]], zeroinitializer
1181 ; AVX512-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 768, i64 784, i64 800, i64 816, i64 832, i64 848, i64 864, i64 880, i64 896, i64 912, i64 928, i64 944, i64 960, i64 976, i64 992, i64 1008>, i32 1
1182 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_3:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP17]], i32 4, <16 x i1> [[TMP16]], <16 x float> undef)
1183 ; AVX512-NEXT: [[TMP18:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_3]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1184 ; AVX512-NEXT: [[TMP19:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 768, i64 784, i64 800, i64 816, i64 832, i64 848, i64 864, i64 880, i64 896, i64 912, i64 928, i64 944, i64 960, i64 976, i64 992, i64 1008>
1185 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP18]], <16 x float*> [[TMP19]], i32 4, <16 x i1> [[TMP16]])
1186 ; AVX512-NEXT: [[TMP20:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 1024, i64 1040, i64 1056, i64 1072, i64 1088, i64 1104, i64 1120, i64 1136, i64 1152, i64 1168, i64 1184, i64 1200, i64 1216, i64 1232, i64 1248, i64 1264>
1187 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_4:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP20]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1188 ; AVX512-NEXT: [[TMP21:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_4]], zeroinitializer
1189 ; AVX512-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 1024, i64 1040, i64 1056, i64 1072, i64 1088, i64 1104, i64 1120, i64 1136, i64 1152, i64 1168, i64 1184, i64 1200, i64 1216, i64 1232, i64 1248, i64 1264>, i32 1
1190 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_4:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP22]], i32 4, <16 x i1> [[TMP21]], <16 x float> undef)
1191 ; AVX512-NEXT: [[TMP23:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_4]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1192 ; AVX512-NEXT: [[TMP24:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 1024, i64 1040, i64 1056, i64 1072, i64 1088, i64 1104, i64 1120, i64 1136, i64 1152, i64 1168, i64 1184, i64 1200, i64 1216, i64 1232, i64 1248, i64 1264>
1193 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP23]], <16 x float*> [[TMP24]], i32 4, <16 x i1> [[TMP21]])
1194 ; AVX512-NEXT: [[TMP25:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 1280, i64 1296, i64 1312, i64 1328, i64 1344, i64 1360, i64 1376, i64 1392, i64 1408, i64 1424, i64 1440, i64 1456, i64 1472, i64 1488, i64 1504, i64 1520>
1195 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_5:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP25]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1196 ; AVX512-NEXT: [[TMP26:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_5]], zeroinitializer
1197 ; AVX512-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 1280, i64 1296, i64 1312, i64 1328, i64 1344, i64 1360, i64 1376, i64 1392, i64 1408, i64 1424, i64 1440, i64 1456, i64 1472, i64 1488, i64 1504, i64 1520>, i32 1
1198 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_5:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP27]], i32 4, <16 x i1> [[TMP26]], <16 x float> undef)
1199 ; AVX512-NEXT: [[TMP28:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_5]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1200 ; AVX512-NEXT: [[TMP29:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 1280, i64 1296, i64 1312, i64 1328, i64 1344, i64 1360, i64 1376, i64 1392, i64 1408, i64 1424, i64 1440, i64 1456, i64 1472, i64 1488, i64 1504, i64 1520>
1201 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP28]], <16 x float*> [[TMP29]], i32 4, <16 x i1> [[TMP26]])
1202 ; AVX512-NEXT: [[TMP30:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 1536, i64 1552, i64 1568, i64 1584, i64 1600, i64 1616, i64 1632, i64 1648, i64 1664, i64 1680, i64 1696, i64 1712, i64 1728, i64 1744, i64 1760, i64 1776>
1203 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_6:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP30]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1204 ; AVX512-NEXT: [[TMP31:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_6]], zeroinitializer
1205 ; AVX512-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 1536, i64 1552, i64 1568, i64 1584, i64 1600, i64 1616, i64 1632, i64 1648, i64 1664, i64 1680, i64 1696, i64 1712, i64 1728, i64 1744, i64 1760, i64 1776>, i32 1
1206 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_6:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP32]], i32 4, <16 x i1> [[TMP31]], <16 x float> undef)
1207 ; AVX512-NEXT: [[TMP33:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_6]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1208 ; AVX512-NEXT: [[TMP34:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 1536, i64 1552, i64 1568, i64 1584, i64 1600, i64 1616, i64 1632, i64 1648, i64 1664, i64 1680, i64 1696, i64 1712, i64 1728, i64 1744, i64 1760, i64 1776>
1209 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP33]], <16 x float*> [[TMP34]], i32 4, <16 x i1> [[TMP31]])
1210 ; AVX512-NEXT: [[TMP35:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 1792, i64 1808, i64 1824, i64 1840, i64 1856, i64 1872, i64 1888, i64 1904, i64 1920, i64 1936, i64 1952, i64 1968, i64 1984, i64 2000, i64 2016, i64 2032>
1211 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_7:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP35]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1212 ; AVX512-NEXT: [[TMP36:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_7]], zeroinitializer
1213 ; AVX512-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 1792, i64 1808, i64 1824, i64 1840, i64 1856, i64 1872, i64 1888, i64 1904, i64 1920, i64 1936, i64 1952, i64 1968, i64 1984, i64 2000, i64 2016, i64 2032>, i32 1
1214 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_7:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP37]], i32 4, <16 x i1> [[TMP36]], <16 x float> undef)
1215 ; AVX512-NEXT: [[TMP38:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_7]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1216 ; AVX512-NEXT: [[TMP39:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 1792, i64 1808, i64 1824, i64 1840, i64 1856, i64 1872, i64 1888, i64 1904, i64 1920, i64 1936, i64 1952, i64 1968, i64 1984, i64 2000, i64 2016, i64 2032>
1217 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP38]], <16 x float*> [[TMP39]], i32 4, <16 x i1> [[TMP36]])
1218 ; AVX512-NEXT: [[TMP40:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 2048, i64 2064, i64 2080, i64 2096, i64 2112, i64 2128, i64 2144, i64 2160, i64 2176, i64 2192, i64 2208, i64 2224, i64 2240, i64 2256, i64 2272, i64 2288>
1219 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_8:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP40]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1220 ; AVX512-NEXT: [[TMP41:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_8]], zeroinitializer
1221 ; AVX512-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 2048, i64 2064, i64 2080, i64 2096, i64 2112, i64 2128, i64 2144, i64 2160, i64 2176, i64 2192, i64 2208, i64 2224, i64 2240, i64 2256, i64 2272, i64 2288>, i32 1
1222 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_8:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP42]], i32 4, <16 x i1> [[TMP41]], <16 x float> undef)
1223 ; AVX512-NEXT: [[TMP43:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_8]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1224 ; AVX512-NEXT: [[TMP44:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 2048, i64 2064, i64 2080, i64 2096, i64 2112, i64 2128, i64 2144, i64 2160, i64 2176, i64 2192, i64 2208, i64 2224, i64 2240, i64 2256, i64 2272, i64 2288>
1225 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP43]], <16 x float*> [[TMP44]], i32 4, <16 x i1> [[TMP41]])
1226 ; AVX512-NEXT: [[TMP45:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 2304, i64 2320, i64 2336, i64 2352, i64 2368, i64 2384, i64 2400, i64 2416, i64 2432, i64 2448, i64 2464, i64 2480, i64 2496, i64 2512, i64 2528, i64 2544>
1227 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_9:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP45]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1228 ; AVX512-NEXT: [[TMP46:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_9]], zeroinitializer
1229 ; AVX512-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 2304, i64 2320, i64 2336, i64 2352, i64 2368, i64 2384, i64 2400, i64 2416, i64 2432, i64 2448, i64 2464, i64 2480, i64 2496, i64 2512, i64 2528, i64 2544>, i32 1
1230 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_9:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP47]], i32 4, <16 x i1> [[TMP46]], <16 x float> undef)
1231 ; AVX512-NEXT: [[TMP48:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_9]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1232 ; AVX512-NEXT: [[TMP49:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 2304, i64 2320, i64 2336, i64 2352, i64 2368, i64 2384, i64 2400, i64 2416, i64 2432, i64 2448, i64 2464, i64 2480, i64 2496, i64 2512, i64 2528, i64 2544>
1233 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP48]], <16 x float*> [[TMP49]], i32 4, <16 x i1> [[TMP46]])
1234 ; AVX512-NEXT: [[TMP50:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 2560, i64 2576, i64 2592, i64 2608, i64 2624, i64 2640, i64 2656, i64 2672, i64 2688, i64 2704, i64 2720, i64 2736, i64 2752, i64 2768, i64 2784, i64 2800>
1235 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_10:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP50]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1236 ; AVX512-NEXT: [[TMP51:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_10]], zeroinitializer
1237 ; AVX512-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 2560, i64 2576, i64 2592, i64 2608, i64 2624, i64 2640, i64 2656, i64 2672, i64 2688, i64 2704, i64 2720, i64 2736, i64 2752, i64 2768, i64 2784, i64 2800>, i32 1
1238 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_10:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP52]], i32 4, <16 x i1> [[TMP51]], <16 x float> undef)
1239 ; AVX512-NEXT: [[TMP53:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_10]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1240 ; AVX512-NEXT: [[TMP54:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 2560, i64 2576, i64 2592, i64 2608, i64 2624, i64 2640, i64 2656, i64 2672, i64 2688, i64 2704, i64 2720, i64 2736, i64 2752, i64 2768, i64 2784, i64 2800>
1241 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP53]], <16 x float*> [[TMP54]], i32 4, <16 x i1> [[TMP51]])
1242 ; AVX512-NEXT: [[TMP55:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 2816, i64 2832, i64 2848, i64 2864, i64 2880, i64 2896, i64 2912, i64 2928, i64 2944, i64 2960, i64 2976, i64 2992, i64 3008, i64 3024, i64 3040, i64 3056>
1243 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_11:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP55]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1244 ; AVX512-NEXT: [[TMP56:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_11]], zeroinitializer
1245 ; AVX512-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 2816, i64 2832, i64 2848, i64 2864, i64 2880, i64 2896, i64 2912, i64 2928, i64 2944, i64 2960, i64 2976, i64 2992, i64 3008, i64 3024, i64 3040, i64 3056>, i32 1
1246 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_11:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP57]], i32 4, <16 x i1> [[TMP56]], <16 x float> undef)
1247 ; AVX512-NEXT: [[TMP58:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_11]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1248 ; AVX512-NEXT: [[TMP59:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 2816, i64 2832, i64 2848, i64 2864, i64 2880, i64 2896, i64 2912, i64 2928, i64 2944, i64 2960, i64 2976, i64 2992, i64 3008, i64 3024, i64 3040, i64 3056>
1249 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP58]], <16 x float*> [[TMP59]], i32 4, <16 x i1> [[TMP56]])
1250 ; AVX512-NEXT: [[TMP60:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 3072, i64 3088, i64 3104, i64 3120, i64 3136, i64 3152, i64 3168, i64 3184, i64 3200, i64 3216, i64 3232, i64 3248, i64 3264, i64 3280, i64 3296, i64 3312>
1251 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_12:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP60]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1252 ; AVX512-NEXT: [[TMP61:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_12]], zeroinitializer
1253 ; AVX512-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 3072, i64 3088, i64 3104, i64 3120, i64 3136, i64 3152, i64 3168, i64 3184, i64 3200, i64 3216, i64 3232, i64 3248, i64 3264, i64 3280, i64 3296, i64 3312>, i32 1
1254 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_12:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP62]], i32 4, <16 x i1> [[TMP61]], <16 x float> undef)
1255 ; AVX512-NEXT: [[TMP63:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_12]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1256 ; AVX512-NEXT: [[TMP64:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 3072, i64 3088, i64 3104, i64 3120, i64 3136, i64 3152, i64 3168, i64 3184, i64 3200, i64 3216, i64 3232, i64 3248, i64 3264, i64 3280, i64 3296, i64 3312>
1257 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP63]], <16 x float*> [[TMP64]], i32 4, <16 x i1> [[TMP61]])
1258 ; AVX512-NEXT: [[TMP65:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 3328, i64 3344, i64 3360, i64 3376, i64 3392, i64 3408, i64 3424, i64 3440, i64 3456, i64 3472, i64 3488, i64 3504, i64 3520, i64 3536, i64 3552, i64 3568>
1259 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_13:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP65]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1260 ; AVX512-NEXT: [[TMP66:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_13]], zeroinitializer
1261 ; AVX512-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 3328, i64 3344, i64 3360, i64 3376, i64 3392, i64 3408, i64 3424, i64 3440, i64 3456, i64 3472, i64 3488, i64 3504, i64 3520, i64 3536, i64 3552, i64 3568>, i32 1
1262 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_13:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP67]], i32 4, <16 x i1> [[TMP66]], <16 x float> undef)
1263 ; AVX512-NEXT: [[TMP68:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_13]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1264 ; AVX512-NEXT: [[TMP69:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 3328, i64 3344, i64 3360, i64 3376, i64 3392, i64 3408, i64 3424, i64 3440, i64 3456, i64 3472, i64 3488, i64 3504, i64 3520, i64 3536, i64 3552, i64 3568>
1265 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP68]], <16 x float*> [[TMP69]], i32 4, <16 x i1> [[TMP66]])
1266 ; AVX512-NEXT: [[TMP70:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 3584, i64 3600, i64 3616, i64 3632, i64 3648, i64 3664, i64 3680, i64 3696, i64 3712, i64 3728, i64 3744, i64 3760, i64 3776, i64 3792, i64 3808, i64 3824>
1267 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_14:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP70]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1268 ; AVX512-NEXT: [[TMP71:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_14]], zeroinitializer
1269 ; AVX512-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 3584, i64 3600, i64 3616, i64 3632, i64 3648, i64 3664, i64 3680, i64 3696, i64 3712, i64 3728, i64 3744, i64 3760, i64 3776, i64 3792, i64 3808, i64 3824>, i32 1
1270 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_14:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP72]], i32 4, <16 x i1> [[TMP71]], <16 x float> undef)
1271 ; AVX512-NEXT: [[TMP73:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_14]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1272 ; AVX512-NEXT: [[TMP74:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 3584, i64 3600, i64 3616, i64 3632, i64 3648, i64 3664, i64 3680, i64 3696, i64 3712, i64 3728, i64 3744, i64 3760, i64 3776, i64 3792, i64 3808, i64 3824>
1273 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP73]], <16 x float*> [[TMP74]], i32 4, <16 x i1> [[TMP71]])
1274 ; AVX512-NEXT: [[TMP75:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 3840, i64 3856, i64 3872, i64 3888, i64 3904, i64 3920, i64 3936, i64 3952, i64 3968, i64 3984, i64 4000, i64 4016, i64 4032, i64 4048, i64 4064, i64 4080>
1275 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_15:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP75]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1276 ; AVX512-NEXT: [[TMP76:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_15]], zeroinitializer
1277 ; AVX512-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <16 x i64> <i64 3840, i64 3856, i64 3872, i64 3888, i64 3904, i64 3920, i64 3936, i64 3952, i64 3968, i64 3984, i64 4000, i64 4016, i64 4032, i64 4048, i64 4064, i64 4080>, i32 1
1278 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_15:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p1f32(<16 x float addrspace(1)*> [[TMP77]], i32 4, <16 x i1> [[TMP76]], <16 x float> undef)
1279 ; AVX512-NEXT: [[TMP78:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_15]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1280 ; AVX512-NEXT: [[TMP79:%.*]] = getelementptr inbounds float, float* [[OUT]], <16 x i64> <i64 3840, i64 3856, i64 3872, i64 3888, i64 3904, i64 3920, i64 3936, i64 3952, i64 3968, i64 3984, i64 4000, i64 4016, i64 4032, i64 4048, i64 4064, i64 4080>
1281 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[TMP78]], <16 x float*> [[TMP79]], i32 4, <16 x i1> [[TMP76]])
1282 ; AVX512-NEXT: ret void
1284 ; FVW2-LABEL: @foo2_addrspace2(
1286 ; FVW2-NEXT: br label [[VECTOR_BODY:%.*]]
1287 ; FVW2: vector.body:
1288 ; FVW2-NEXT: [[INDEX6:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE30:%.*]] ]
1289 ; FVW2-NEXT: [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 16>, [[ENTRY]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_STORE_CONTINUE30]] ]
1290 ; FVW2-NEXT: [[STEP_ADD:%.*]] = add <2 x i64> [[VEC_IND]], <i64 32, i64 32>
1291 ; FVW2-NEXT: [[STEP_ADD7:%.*]] = add <2 x i64> [[VEC_IND]], <i64 64, i64 64>
1292 ; FVW2-NEXT: [[STEP_ADD8:%.*]] = add <2 x i64> [[VEC_IND]], <i64 96, i64 96>
1293 ; FVW2-NEXT: [[OFFSET_IDX:%.*]] = shl i64 [[INDEX6]], 4
1294 ; FVW2-NEXT: [[TMP0:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER:%.*]], <2 x i64> [[VEC_IND]]
1295 ; FVW2-NEXT: [[TMP1:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <2 x i64> [[STEP_ADD]]
1296 ; FVW2-NEXT: [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <2 x i64> [[STEP_ADD7]]
1297 ; FVW2-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <2 x i64> [[STEP_ADD8]]
1298 ; FVW2-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <2 x i32> @llvm.masked.gather.v2i32.v2p0i32(<2 x i32*> [[TMP0]], i32 4, <2 x i1> <i1 true, i1 true>, <2 x i32> undef)
1299 ; FVW2-NEXT: [[WIDE_MASKED_GATHER10:%.*]] = call <2 x i32> @llvm.masked.gather.v2i32.v2p0i32(<2 x i32*> [[TMP1]], i32 4, <2 x i1> <i1 true, i1 true>, <2 x i32> undef)
1300 ; FVW2-NEXT: [[WIDE_MASKED_GATHER11:%.*]] = call <2 x i32> @llvm.masked.gather.v2i32.v2p0i32(<2 x i32*> [[TMP2]], i32 4, <2 x i1> <i1 true, i1 true>, <2 x i32> undef)
1301 ; FVW2-NEXT: [[WIDE_MASKED_GATHER12:%.*]] = call <2 x i32> @llvm.masked.gather.v2i32.v2p0i32(<2 x i32*> [[TMP3]], i32 4, <2 x i1> <i1 true, i1 true>, <2 x i32> undef)
1302 ; FVW2-NEXT: [[TMP4:%.*]] = icmp sgt <2 x i32> [[WIDE_MASKED_GATHER]], zeroinitializer
1303 ; FVW2-NEXT: [[TMP5:%.*]] = icmp sgt <2 x i32> [[WIDE_MASKED_GATHER10]], zeroinitializer
1304 ; FVW2-NEXT: [[TMP6:%.*]] = icmp sgt <2 x i32> [[WIDE_MASKED_GATHER11]], zeroinitializer
1305 ; FVW2-NEXT: [[TMP7:%.*]] = icmp sgt <2 x i32> [[WIDE_MASKED_GATHER12]], zeroinitializer
1306 ; FVW2-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT_IN:%.*]], [[STRUCT_IN]] addrspace(1)* [[IN:%.*]], <2 x i64> [[VEC_IND]], i32 1
1307 ; FVW2-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <2 x i64> [[STEP_ADD]], i32 1
1308 ; FVW2-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <2 x i64> [[STEP_ADD7]], i32 1
1309 ; FVW2-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_IN]], [[STRUCT_IN]] addrspace(1)* [[IN]], <2 x i64> [[STEP_ADD8]], i32 1
1310 ; FVW2-NEXT: [[WIDE_MASKED_GATHER13:%.*]] = call <2 x float> @llvm.masked.gather.v2f32.v2p1f32(<2 x float addrspace(1)*> [[TMP8]], i32 4, <2 x i1> [[TMP4]], <2 x float> undef)
1311 ; FVW2-NEXT: [[WIDE_MASKED_GATHER14:%.*]] = call <2 x float> @llvm.masked.gather.v2f32.v2p1f32(<2 x float addrspace(1)*> [[TMP9]], i32 4, <2 x i1> [[TMP5]], <2 x float> undef)
1312 ; FVW2-NEXT: [[WIDE_MASKED_GATHER15:%.*]] = call <2 x float> @llvm.masked.gather.v2f32.v2p1f32(<2 x float addrspace(1)*> [[TMP10]], i32 4, <2 x i1> [[TMP6]], <2 x float> undef)
1313 ; FVW2-NEXT: [[WIDE_MASKED_GATHER16:%.*]] = call <2 x float> @llvm.masked.gather.v2f32.v2p1f32(<2 x float addrspace(1)*> [[TMP11]], i32 4, <2 x i1> [[TMP7]], <2 x float> undef)
1314 ; FVW2-NEXT: [[TMP12:%.*]] = fadd <2 x float> [[WIDE_MASKED_GATHER13]], <float 5.000000e-01, float 5.000000e-01>
1315 ; FVW2-NEXT: [[TMP13:%.*]] = fadd <2 x float> [[WIDE_MASKED_GATHER14]], <float 5.000000e-01, float 5.000000e-01>
1316 ; FVW2-NEXT: [[TMP14:%.*]] = fadd <2 x float> [[WIDE_MASKED_GATHER15]], <float 5.000000e-01, float 5.000000e-01>
1317 ; FVW2-NEXT: [[TMP15:%.*]] = fadd <2 x float> [[WIDE_MASKED_GATHER16]], <float 5.000000e-01, float 5.000000e-01>
1318 ; FVW2-NEXT: [[TMP16:%.*]] = extractelement <2 x i1> [[TMP4]], i32 0
1319 ; FVW2-NEXT: br i1 [[TMP16]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
1320 ; FVW2: pred.store.if:
1321 ; FVW2-NEXT: [[TMP17:%.*]] = getelementptr inbounds float, float* [[OUT:%.*]], i64 [[OFFSET_IDX]]
1322 ; FVW2-NEXT: [[TMP18:%.*]] = extractelement <2 x float> [[TMP12]], i32 0
1323 ; FVW2-NEXT: store float [[TMP18]], float* [[TMP17]], align 4
1324 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE]]
1325 ; FVW2: pred.store.continue:
1326 ; FVW2-NEXT: [[TMP19:%.*]] = extractelement <2 x i1> [[TMP4]], i32 1
1327 ; FVW2-NEXT: br i1 [[TMP19]], label [[PRED_STORE_IF17:%.*]], label [[PRED_STORE_CONTINUE18:%.*]]
1328 ; FVW2: pred.store.if17:
1329 ; FVW2-NEXT: [[TMP20:%.*]] = or i64 [[OFFSET_IDX]], 16
1330 ; FVW2-NEXT: [[TMP21:%.*]] = getelementptr inbounds float, float* [[OUT]], i64 [[TMP20]]
1331 ; FVW2-NEXT: [[TMP22:%.*]] = extractelement <2 x float> [[TMP12]], i32 1
1332 ; FVW2-NEXT: store float [[TMP22]], float* [[TMP21]], align 4
1333 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE18]]
1334 ; FVW2: pred.store.continue18:
1335 ; FVW2-NEXT: [[TMP23:%.*]] = extractelement <2 x i1> [[TMP5]], i32 0
1336 ; FVW2-NEXT: br i1 [[TMP23]], label [[PRED_STORE_IF19:%.*]], label [[PRED_STORE_CONTINUE20:%.*]]
1337 ; FVW2: pred.store.if19:
1338 ; FVW2-NEXT: [[TMP24:%.*]] = or i64 [[OFFSET_IDX]], 32
1339 ; FVW2-NEXT: [[TMP25:%.*]] = getelementptr inbounds float, float* [[OUT]], i64 [[TMP24]]
1340 ; FVW2-NEXT: [[TMP26:%.*]] = extractelement <2 x float> [[TMP13]], i32 0
1341 ; FVW2-NEXT: store float [[TMP26]], float* [[TMP25]], align 4
1342 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE20]]
1343 ; FVW2: pred.store.continue20:
1344 ; FVW2-NEXT: [[TMP27:%.*]] = extractelement <2 x i1> [[TMP5]], i32 1
1345 ; FVW2-NEXT: br i1 [[TMP27]], label [[PRED_STORE_IF21:%.*]], label [[PRED_STORE_CONTINUE22:%.*]]
1346 ; FVW2: pred.store.if21:
1347 ; FVW2-NEXT: [[TMP28:%.*]] = or i64 [[OFFSET_IDX]], 48
1348 ; FVW2-NEXT: [[TMP29:%.*]] = getelementptr inbounds float, float* [[OUT]], i64 [[TMP28]]
1349 ; FVW2-NEXT: [[TMP30:%.*]] = extractelement <2 x float> [[TMP13]], i32 1
1350 ; FVW2-NEXT: store float [[TMP30]], float* [[TMP29]], align 4
1351 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE22]]
1352 ; FVW2: pred.store.continue22:
1353 ; FVW2-NEXT: [[TMP31:%.*]] = extractelement <2 x i1> [[TMP6]], i32 0
1354 ; FVW2-NEXT: br i1 [[TMP31]], label [[PRED_STORE_IF23:%.*]], label [[PRED_STORE_CONTINUE24:%.*]]
1355 ; FVW2: pred.store.if23:
1356 ; FVW2-NEXT: [[TMP32:%.*]] = or i64 [[OFFSET_IDX]], 64
1357 ; FVW2-NEXT: [[TMP33:%.*]] = getelementptr inbounds float, float* [[OUT]], i64 [[TMP32]]
1358 ; FVW2-NEXT: [[TMP34:%.*]] = extractelement <2 x float> [[TMP14]], i32 0
1359 ; FVW2-NEXT: store float [[TMP34]], float* [[TMP33]], align 4
1360 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE24]]
1361 ; FVW2: pred.store.continue24:
1362 ; FVW2-NEXT: [[TMP35:%.*]] = extractelement <2 x i1> [[TMP6]], i32 1
1363 ; FVW2-NEXT: br i1 [[TMP35]], label [[PRED_STORE_IF25:%.*]], label [[PRED_STORE_CONTINUE26:%.*]]
1364 ; FVW2: pred.store.if25:
1365 ; FVW2-NEXT: [[TMP36:%.*]] = or i64 [[OFFSET_IDX]], 80
1366 ; FVW2-NEXT: [[TMP37:%.*]] = getelementptr inbounds float, float* [[OUT]], i64 [[TMP36]]
1367 ; FVW2-NEXT: [[TMP38:%.*]] = extractelement <2 x float> [[TMP14]], i32 1
1368 ; FVW2-NEXT: store float [[TMP38]], float* [[TMP37]], align 4
1369 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE26]]
1370 ; FVW2: pred.store.continue26:
1371 ; FVW2-NEXT: [[TMP39:%.*]] = extractelement <2 x i1> [[TMP7]], i32 0
1372 ; FVW2-NEXT: br i1 [[TMP39]], label [[PRED_STORE_IF27:%.*]], label [[PRED_STORE_CONTINUE28:%.*]]
1373 ; FVW2: pred.store.if27:
1374 ; FVW2-NEXT: [[TMP40:%.*]] = or i64 [[OFFSET_IDX]], 96
1375 ; FVW2-NEXT: [[TMP41:%.*]] = getelementptr inbounds float, float* [[OUT]], i64 [[TMP40]]
1376 ; FVW2-NEXT: [[TMP42:%.*]] = extractelement <2 x float> [[TMP15]], i32 0
1377 ; FVW2-NEXT: store float [[TMP42]], float* [[TMP41]], align 4
1378 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE28]]
1379 ; FVW2: pred.store.continue28:
1380 ; FVW2-NEXT: [[TMP43:%.*]] = extractelement <2 x i1> [[TMP7]], i32 1
1381 ; FVW2-NEXT: br i1 [[TMP43]], label [[PRED_STORE_IF29:%.*]], label [[PRED_STORE_CONTINUE30]]
1382 ; FVW2: pred.store.if29:
1383 ; FVW2-NEXT: [[TMP44:%.*]] = or i64 [[OFFSET_IDX]], 112
1384 ; FVW2-NEXT: [[TMP45:%.*]] = getelementptr inbounds float, float* [[OUT]], i64 [[TMP44]]
1385 ; FVW2-NEXT: [[TMP46:%.*]] = extractelement <2 x float> [[TMP15]], i32 1
1386 ; FVW2-NEXT: store float [[TMP46]], float* [[TMP45]], align 4
1387 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE30]]
1388 ; FVW2: pred.store.continue30:
1389 ; FVW2-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX6]], 8
1390 ; FVW2-NEXT: [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], <i64 128, i64 128>
1391 ; FVW2-NEXT: [[TMP47:%.*]] = icmp eq i64 [[INDEX_NEXT]], 256
1392 ; FVW2-NEXT: br i1 [[TMP47]], label [[FOR_END:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
1394 ; FVW2-NEXT: ret void
1397 %in.addr = alloca %struct.In addrspace(1)*, align 8
1398 %out.addr = alloca float addrspace(0)*, align 8
1399 %trigger.addr = alloca i32*, align 8
1400 %index.addr = alloca i32*, align 8
1401 %i = alloca i32, align 4
1402 store %struct.In addrspace(1)* %in, %struct.In addrspace(1)** %in.addr, align 8
1403 store float addrspace(0)* %out, float addrspace(0)** %out.addr, align 8
1404 store i32* %trigger, i32** %trigger.addr, align 8
1405 store i32* %index, i32** %index.addr, align 8
1406 store i32 0, i32* %i, align 4
1409 for.cond: ; preds = %for.inc, %entry
1410 %0 = load i32, i32* %i, align 4
1411 %cmp = icmp slt i32 %0, 4096
1412 br i1 %cmp, label %for.body, label %for.end
1414 for.body: ; preds = %for.cond
1415 %1 = load i32, i32* %i, align 4
1416 %idxprom = sext i32 %1 to i64
1417 %2 = load i32*, i32** %trigger.addr, align 8
1418 %arrayidx = getelementptr inbounds i32, i32* %2, i64 %idxprom
1419 %3 = load i32, i32* %arrayidx, align 4
1420 %cmp1 = icmp sgt i32 %3, 0
1421 br i1 %cmp1, label %if.then, label %if.end
1423 if.then: ; preds = %for.body
1424 %4 = load i32, i32* %i, align 4
1425 %idxprom2 = sext i32 %4 to i64
1426 %5 = load %struct.In addrspace(1)*, %struct.In addrspace(1)** %in.addr, align 8
1427 %arrayidx3 = getelementptr inbounds %struct.In, %struct.In addrspace(1)* %5, i64 %idxprom2
1428 %b = getelementptr inbounds %struct.In, %struct.In addrspace(1)* %arrayidx3, i32 0, i32 1
1429 %6 = load float, float addrspace(1)* %b, align 4
1430 %add = fadd float %6, 5.000000e-01
1431 %7 = load i32, i32* %i, align 4
1432 %idxprom4 = sext i32 %7 to i64
1433 %8 = load float addrspace(0)*, float addrspace(0)** %out.addr, align 8
1434 %arrayidx5 = getelementptr inbounds float, float addrspace(0)* %8, i64 %idxprom4
1435 store float %add, float addrspace(0)* %arrayidx5, align 4
1438 if.end: ; preds = %if.then, %for.body
1441 for.inc: ; preds = %if.end
1442 %9 = load i32, i32* %i, align 4
1443 %inc = add nsw i32 %9, 16
1444 store i32 %inc, i32* %i, align 4
1447 for.end: ; preds = %for.cond
1451 ; Same as foo2_addrspace but here only the output has the non-default address space.
1453 define void @foo2_addrspace3(%struct.In addrspace(0)* noalias %in, float addrspace(1)* noalias %out, i32* noalias %trigger, i32* noalias %index) {
1454 ; AVX512-LABEL: @foo2_addrspace3(
1455 ; AVX512-NEXT: entry:
1456 ; AVX512-NEXT: [[TMP0:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER:%.*]], <16 x i64> <i64 0, i64 16, i64 32, i64 48, i64 64, i64 80, i64 96, i64 112, i64 128, i64 144, i64 160, i64 176, i64 192, i64 208, i64 224, i64 240>
1457 ; AVX512-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP0]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1458 ; AVX512-NEXT: [[TMP1:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER]], zeroinitializer
1459 ; AVX512-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT_IN:%.*]], %struct.In* [[IN:%.*]], <16 x i64> <i64 0, i64 16, i64 32, i64 48, i64 64, i64 80, i64 96, i64 112, i64 128, i64 144, i64 160, i64 176, i64 192, i64 208, i64 224, i64 240>, i32 1
1460 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP2]], i32 4, <16 x i1> [[TMP1]], <16 x float> undef)
1461 ; AVX512-NEXT: [[TMP3:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1462 ; AVX512-NEXT: [[TMP4:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT:%.*]], <16 x i64> <i64 0, i64 16, i64 32, i64 48, i64 64, i64 80, i64 96, i64 112, i64 128, i64 144, i64 160, i64 176, i64 192, i64 208, i64 224, i64 240>
1463 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP3]], <16 x float addrspace(1)*> [[TMP4]], i32 4, <16 x i1> [[TMP1]])
1464 ; AVX512-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 256, i64 272, i64 288, i64 304, i64 320, i64 336, i64 352, i64 368, i64 384, i64 400, i64 416, i64 432, i64 448, i64 464, i64 480, i64 496>
1465 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_1:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP5]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1466 ; AVX512-NEXT: [[TMP6:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_1]], zeroinitializer
1467 ; AVX512-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 256, i64 272, i64 288, i64 304, i64 320, i64 336, i64 352, i64 368, i64 384, i64 400, i64 416, i64 432, i64 448, i64 464, i64 480, i64 496>, i32 1
1468 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_1:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP7]], i32 4, <16 x i1> [[TMP6]], <16 x float> undef)
1469 ; AVX512-NEXT: [[TMP8:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_1]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1470 ; AVX512-NEXT: [[TMP9:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 256, i64 272, i64 288, i64 304, i64 320, i64 336, i64 352, i64 368, i64 384, i64 400, i64 416, i64 432, i64 448, i64 464, i64 480, i64 496>
1471 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP8]], <16 x float addrspace(1)*> [[TMP9]], i32 4, <16 x i1> [[TMP6]])
1472 ; AVX512-NEXT: [[TMP10:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 512, i64 528, i64 544, i64 560, i64 576, i64 592, i64 608, i64 624, i64 640, i64 656, i64 672, i64 688, i64 704, i64 720, i64 736, i64 752>
1473 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_2:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP10]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1474 ; AVX512-NEXT: [[TMP11:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_2]], zeroinitializer
1475 ; AVX512-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 512, i64 528, i64 544, i64 560, i64 576, i64 592, i64 608, i64 624, i64 640, i64 656, i64 672, i64 688, i64 704, i64 720, i64 736, i64 752>, i32 1
1476 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_2:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP12]], i32 4, <16 x i1> [[TMP11]], <16 x float> undef)
1477 ; AVX512-NEXT: [[TMP13:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_2]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1478 ; AVX512-NEXT: [[TMP14:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 512, i64 528, i64 544, i64 560, i64 576, i64 592, i64 608, i64 624, i64 640, i64 656, i64 672, i64 688, i64 704, i64 720, i64 736, i64 752>
1479 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP13]], <16 x float addrspace(1)*> [[TMP14]], i32 4, <16 x i1> [[TMP11]])
1480 ; AVX512-NEXT: [[TMP15:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 768, i64 784, i64 800, i64 816, i64 832, i64 848, i64 864, i64 880, i64 896, i64 912, i64 928, i64 944, i64 960, i64 976, i64 992, i64 1008>
1481 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_3:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP15]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1482 ; AVX512-NEXT: [[TMP16:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_3]], zeroinitializer
1483 ; AVX512-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 768, i64 784, i64 800, i64 816, i64 832, i64 848, i64 864, i64 880, i64 896, i64 912, i64 928, i64 944, i64 960, i64 976, i64 992, i64 1008>, i32 1
1484 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_3:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP17]], i32 4, <16 x i1> [[TMP16]], <16 x float> undef)
1485 ; AVX512-NEXT: [[TMP18:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_3]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1486 ; AVX512-NEXT: [[TMP19:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 768, i64 784, i64 800, i64 816, i64 832, i64 848, i64 864, i64 880, i64 896, i64 912, i64 928, i64 944, i64 960, i64 976, i64 992, i64 1008>
1487 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP18]], <16 x float addrspace(1)*> [[TMP19]], i32 4, <16 x i1> [[TMP16]])
1488 ; AVX512-NEXT: [[TMP20:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 1024, i64 1040, i64 1056, i64 1072, i64 1088, i64 1104, i64 1120, i64 1136, i64 1152, i64 1168, i64 1184, i64 1200, i64 1216, i64 1232, i64 1248, i64 1264>
1489 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_4:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP20]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1490 ; AVX512-NEXT: [[TMP21:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_4]], zeroinitializer
1491 ; AVX512-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 1024, i64 1040, i64 1056, i64 1072, i64 1088, i64 1104, i64 1120, i64 1136, i64 1152, i64 1168, i64 1184, i64 1200, i64 1216, i64 1232, i64 1248, i64 1264>, i32 1
1492 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_4:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP22]], i32 4, <16 x i1> [[TMP21]], <16 x float> undef)
1493 ; AVX512-NEXT: [[TMP23:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_4]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1494 ; AVX512-NEXT: [[TMP24:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 1024, i64 1040, i64 1056, i64 1072, i64 1088, i64 1104, i64 1120, i64 1136, i64 1152, i64 1168, i64 1184, i64 1200, i64 1216, i64 1232, i64 1248, i64 1264>
1495 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP23]], <16 x float addrspace(1)*> [[TMP24]], i32 4, <16 x i1> [[TMP21]])
1496 ; AVX512-NEXT: [[TMP25:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 1280, i64 1296, i64 1312, i64 1328, i64 1344, i64 1360, i64 1376, i64 1392, i64 1408, i64 1424, i64 1440, i64 1456, i64 1472, i64 1488, i64 1504, i64 1520>
1497 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_5:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP25]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1498 ; AVX512-NEXT: [[TMP26:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_5]], zeroinitializer
1499 ; AVX512-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 1280, i64 1296, i64 1312, i64 1328, i64 1344, i64 1360, i64 1376, i64 1392, i64 1408, i64 1424, i64 1440, i64 1456, i64 1472, i64 1488, i64 1504, i64 1520>, i32 1
1500 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_5:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP27]], i32 4, <16 x i1> [[TMP26]], <16 x float> undef)
1501 ; AVX512-NEXT: [[TMP28:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_5]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1502 ; AVX512-NEXT: [[TMP29:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 1280, i64 1296, i64 1312, i64 1328, i64 1344, i64 1360, i64 1376, i64 1392, i64 1408, i64 1424, i64 1440, i64 1456, i64 1472, i64 1488, i64 1504, i64 1520>
1503 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP28]], <16 x float addrspace(1)*> [[TMP29]], i32 4, <16 x i1> [[TMP26]])
1504 ; AVX512-NEXT: [[TMP30:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 1536, i64 1552, i64 1568, i64 1584, i64 1600, i64 1616, i64 1632, i64 1648, i64 1664, i64 1680, i64 1696, i64 1712, i64 1728, i64 1744, i64 1760, i64 1776>
1505 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_6:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP30]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1506 ; AVX512-NEXT: [[TMP31:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_6]], zeroinitializer
1507 ; AVX512-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 1536, i64 1552, i64 1568, i64 1584, i64 1600, i64 1616, i64 1632, i64 1648, i64 1664, i64 1680, i64 1696, i64 1712, i64 1728, i64 1744, i64 1760, i64 1776>, i32 1
1508 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_6:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP32]], i32 4, <16 x i1> [[TMP31]], <16 x float> undef)
1509 ; AVX512-NEXT: [[TMP33:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_6]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1510 ; AVX512-NEXT: [[TMP34:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 1536, i64 1552, i64 1568, i64 1584, i64 1600, i64 1616, i64 1632, i64 1648, i64 1664, i64 1680, i64 1696, i64 1712, i64 1728, i64 1744, i64 1760, i64 1776>
1511 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP33]], <16 x float addrspace(1)*> [[TMP34]], i32 4, <16 x i1> [[TMP31]])
1512 ; AVX512-NEXT: [[TMP35:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 1792, i64 1808, i64 1824, i64 1840, i64 1856, i64 1872, i64 1888, i64 1904, i64 1920, i64 1936, i64 1952, i64 1968, i64 1984, i64 2000, i64 2016, i64 2032>
1513 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_7:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP35]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1514 ; AVX512-NEXT: [[TMP36:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_7]], zeroinitializer
1515 ; AVX512-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 1792, i64 1808, i64 1824, i64 1840, i64 1856, i64 1872, i64 1888, i64 1904, i64 1920, i64 1936, i64 1952, i64 1968, i64 1984, i64 2000, i64 2016, i64 2032>, i32 1
1516 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_7:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP37]], i32 4, <16 x i1> [[TMP36]], <16 x float> undef)
1517 ; AVX512-NEXT: [[TMP38:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_7]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1518 ; AVX512-NEXT: [[TMP39:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 1792, i64 1808, i64 1824, i64 1840, i64 1856, i64 1872, i64 1888, i64 1904, i64 1920, i64 1936, i64 1952, i64 1968, i64 1984, i64 2000, i64 2016, i64 2032>
1519 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP38]], <16 x float addrspace(1)*> [[TMP39]], i32 4, <16 x i1> [[TMP36]])
1520 ; AVX512-NEXT: [[TMP40:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 2048, i64 2064, i64 2080, i64 2096, i64 2112, i64 2128, i64 2144, i64 2160, i64 2176, i64 2192, i64 2208, i64 2224, i64 2240, i64 2256, i64 2272, i64 2288>
1521 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_8:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP40]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1522 ; AVX512-NEXT: [[TMP41:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_8]], zeroinitializer
1523 ; AVX512-NEXT: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 2048, i64 2064, i64 2080, i64 2096, i64 2112, i64 2128, i64 2144, i64 2160, i64 2176, i64 2192, i64 2208, i64 2224, i64 2240, i64 2256, i64 2272, i64 2288>, i32 1
1524 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_8:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP42]], i32 4, <16 x i1> [[TMP41]], <16 x float> undef)
1525 ; AVX512-NEXT: [[TMP43:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_8]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1526 ; AVX512-NEXT: [[TMP44:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 2048, i64 2064, i64 2080, i64 2096, i64 2112, i64 2128, i64 2144, i64 2160, i64 2176, i64 2192, i64 2208, i64 2224, i64 2240, i64 2256, i64 2272, i64 2288>
1527 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP43]], <16 x float addrspace(1)*> [[TMP44]], i32 4, <16 x i1> [[TMP41]])
1528 ; AVX512-NEXT: [[TMP45:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 2304, i64 2320, i64 2336, i64 2352, i64 2368, i64 2384, i64 2400, i64 2416, i64 2432, i64 2448, i64 2464, i64 2480, i64 2496, i64 2512, i64 2528, i64 2544>
1529 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_9:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP45]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1530 ; AVX512-NEXT: [[TMP46:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_9]], zeroinitializer
1531 ; AVX512-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 2304, i64 2320, i64 2336, i64 2352, i64 2368, i64 2384, i64 2400, i64 2416, i64 2432, i64 2448, i64 2464, i64 2480, i64 2496, i64 2512, i64 2528, i64 2544>, i32 1
1532 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_9:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP47]], i32 4, <16 x i1> [[TMP46]], <16 x float> undef)
1533 ; AVX512-NEXT: [[TMP48:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_9]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1534 ; AVX512-NEXT: [[TMP49:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 2304, i64 2320, i64 2336, i64 2352, i64 2368, i64 2384, i64 2400, i64 2416, i64 2432, i64 2448, i64 2464, i64 2480, i64 2496, i64 2512, i64 2528, i64 2544>
1535 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP48]], <16 x float addrspace(1)*> [[TMP49]], i32 4, <16 x i1> [[TMP46]])
1536 ; AVX512-NEXT: [[TMP50:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 2560, i64 2576, i64 2592, i64 2608, i64 2624, i64 2640, i64 2656, i64 2672, i64 2688, i64 2704, i64 2720, i64 2736, i64 2752, i64 2768, i64 2784, i64 2800>
1537 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_10:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP50]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1538 ; AVX512-NEXT: [[TMP51:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_10]], zeroinitializer
1539 ; AVX512-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 2560, i64 2576, i64 2592, i64 2608, i64 2624, i64 2640, i64 2656, i64 2672, i64 2688, i64 2704, i64 2720, i64 2736, i64 2752, i64 2768, i64 2784, i64 2800>, i32 1
1540 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_10:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP52]], i32 4, <16 x i1> [[TMP51]], <16 x float> undef)
1541 ; AVX512-NEXT: [[TMP53:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_10]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1542 ; AVX512-NEXT: [[TMP54:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 2560, i64 2576, i64 2592, i64 2608, i64 2624, i64 2640, i64 2656, i64 2672, i64 2688, i64 2704, i64 2720, i64 2736, i64 2752, i64 2768, i64 2784, i64 2800>
1543 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP53]], <16 x float addrspace(1)*> [[TMP54]], i32 4, <16 x i1> [[TMP51]])
1544 ; AVX512-NEXT: [[TMP55:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 2816, i64 2832, i64 2848, i64 2864, i64 2880, i64 2896, i64 2912, i64 2928, i64 2944, i64 2960, i64 2976, i64 2992, i64 3008, i64 3024, i64 3040, i64 3056>
1545 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_11:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP55]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1546 ; AVX512-NEXT: [[TMP56:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_11]], zeroinitializer
1547 ; AVX512-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 2816, i64 2832, i64 2848, i64 2864, i64 2880, i64 2896, i64 2912, i64 2928, i64 2944, i64 2960, i64 2976, i64 2992, i64 3008, i64 3024, i64 3040, i64 3056>, i32 1
1548 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_11:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP57]], i32 4, <16 x i1> [[TMP56]], <16 x float> undef)
1549 ; AVX512-NEXT: [[TMP58:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_11]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1550 ; AVX512-NEXT: [[TMP59:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 2816, i64 2832, i64 2848, i64 2864, i64 2880, i64 2896, i64 2912, i64 2928, i64 2944, i64 2960, i64 2976, i64 2992, i64 3008, i64 3024, i64 3040, i64 3056>
1551 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP58]], <16 x float addrspace(1)*> [[TMP59]], i32 4, <16 x i1> [[TMP56]])
1552 ; AVX512-NEXT: [[TMP60:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 3072, i64 3088, i64 3104, i64 3120, i64 3136, i64 3152, i64 3168, i64 3184, i64 3200, i64 3216, i64 3232, i64 3248, i64 3264, i64 3280, i64 3296, i64 3312>
1553 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_12:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP60]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1554 ; AVX512-NEXT: [[TMP61:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_12]], zeroinitializer
1555 ; AVX512-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 3072, i64 3088, i64 3104, i64 3120, i64 3136, i64 3152, i64 3168, i64 3184, i64 3200, i64 3216, i64 3232, i64 3248, i64 3264, i64 3280, i64 3296, i64 3312>, i32 1
1556 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_12:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP62]], i32 4, <16 x i1> [[TMP61]], <16 x float> undef)
1557 ; AVX512-NEXT: [[TMP63:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_12]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1558 ; AVX512-NEXT: [[TMP64:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 3072, i64 3088, i64 3104, i64 3120, i64 3136, i64 3152, i64 3168, i64 3184, i64 3200, i64 3216, i64 3232, i64 3248, i64 3264, i64 3280, i64 3296, i64 3312>
1559 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP63]], <16 x float addrspace(1)*> [[TMP64]], i32 4, <16 x i1> [[TMP61]])
1560 ; AVX512-NEXT: [[TMP65:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 3328, i64 3344, i64 3360, i64 3376, i64 3392, i64 3408, i64 3424, i64 3440, i64 3456, i64 3472, i64 3488, i64 3504, i64 3520, i64 3536, i64 3552, i64 3568>
1561 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_13:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP65]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1562 ; AVX512-NEXT: [[TMP66:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_13]], zeroinitializer
1563 ; AVX512-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 3328, i64 3344, i64 3360, i64 3376, i64 3392, i64 3408, i64 3424, i64 3440, i64 3456, i64 3472, i64 3488, i64 3504, i64 3520, i64 3536, i64 3552, i64 3568>, i32 1
1564 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_13:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP67]], i32 4, <16 x i1> [[TMP66]], <16 x float> undef)
1565 ; AVX512-NEXT: [[TMP68:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_13]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1566 ; AVX512-NEXT: [[TMP69:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 3328, i64 3344, i64 3360, i64 3376, i64 3392, i64 3408, i64 3424, i64 3440, i64 3456, i64 3472, i64 3488, i64 3504, i64 3520, i64 3536, i64 3552, i64 3568>
1567 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP68]], <16 x float addrspace(1)*> [[TMP69]], i32 4, <16 x i1> [[TMP66]])
1568 ; AVX512-NEXT: [[TMP70:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 3584, i64 3600, i64 3616, i64 3632, i64 3648, i64 3664, i64 3680, i64 3696, i64 3712, i64 3728, i64 3744, i64 3760, i64 3776, i64 3792, i64 3808, i64 3824>
1569 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_14:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP70]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1570 ; AVX512-NEXT: [[TMP71:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_14]], zeroinitializer
1571 ; AVX512-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 3584, i64 3600, i64 3616, i64 3632, i64 3648, i64 3664, i64 3680, i64 3696, i64 3712, i64 3728, i64 3744, i64 3760, i64 3776, i64 3792, i64 3808, i64 3824>, i32 1
1572 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_14:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP72]], i32 4, <16 x i1> [[TMP71]], <16 x float> undef)
1573 ; AVX512-NEXT: [[TMP73:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_14]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1574 ; AVX512-NEXT: [[TMP74:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 3584, i64 3600, i64 3616, i64 3632, i64 3648, i64 3664, i64 3680, i64 3696, i64 3712, i64 3728, i64 3744, i64 3760, i64 3776, i64 3792, i64 3808, i64 3824>
1575 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP73]], <16 x float addrspace(1)*> [[TMP74]], i32 4, <16 x i1> [[TMP71]])
1576 ; AVX512-NEXT: [[TMP75:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <16 x i64> <i64 3840, i64 3856, i64 3872, i64 3888, i64 3904, i64 3920, i64 3936, i64 3952, i64 3968, i64 3984, i64 4000, i64 4016, i64 4032, i64 4048, i64 4064, i64 4080>
1577 ; AVX512-NEXT: [[WIDE_MASKED_GATHER_15:%.*]] = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> [[TMP75]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i32> undef)
1578 ; AVX512-NEXT: [[TMP76:%.*]] = icmp sgt <16 x i32> [[WIDE_MASKED_GATHER_15]], zeroinitializer
1579 ; AVX512-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <16 x i64> <i64 3840, i64 3856, i64 3872, i64 3888, i64 3904, i64 3920, i64 3936, i64 3952, i64 3968, i64 3984, i64 4000, i64 4016, i64 4032, i64 4048, i64 4064, i64 4080>, i32 1
1580 ; AVX512-NEXT: [[WIDE_MASKED_GATHER7_15:%.*]] = call <16 x float> @llvm.masked.gather.v16f32.v16p0f32(<16 x float*> [[TMP77]], i32 4, <16 x i1> [[TMP76]], <16 x float> undef)
1581 ; AVX512-NEXT: [[TMP78:%.*]] = fadd <16 x float> [[WIDE_MASKED_GATHER7_15]], <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01>
1582 ; AVX512-NEXT: [[TMP79:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], <16 x i64> <i64 3840, i64 3856, i64 3872, i64 3888, i64 3904, i64 3920, i64 3936, i64 3952, i64 3968, i64 3984, i64 4000, i64 4016, i64 4032, i64 4048, i64 4064, i64 4080>
1583 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p1f32(<16 x float> [[TMP78]], <16 x float addrspace(1)*> [[TMP79]], i32 4, <16 x i1> [[TMP76]])
1584 ; AVX512-NEXT: ret void
1586 ; FVW2-LABEL: @foo2_addrspace3(
1588 ; FVW2-NEXT: br label [[VECTOR_BODY:%.*]]
1589 ; FVW2: vector.body:
1590 ; FVW2-NEXT: [[INDEX6:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE30:%.*]] ]
1591 ; FVW2-NEXT: [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 16>, [[ENTRY]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_STORE_CONTINUE30]] ]
1592 ; FVW2-NEXT: [[STEP_ADD:%.*]] = add <2 x i64> [[VEC_IND]], <i64 32, i64 32>
1593 ; FVW2-NEXT: [[STEP_ADD7:%.*]] = add <2 x i64> [[VEC_IND]], <i64 64, i64 64>
1594 ; FVW2-NEXT: [[STEP_ADD8:%.*]] = add <2 x i64> [[VEC_IND]], <i64 96, i64 96>
1595 ; FVW2-NEXT: [[OFFSET_IDX:%.*]] = shl i64 [[INDEX6]], 4
1596 ; FVW2-NEXT: [[TMP0:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER:%.*]], <2 x i64> [[VEC_IND]]
1597 ; FVW2-NEXT: [[TMP1:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <2 x i64> [[STEP_ADD]]
1598 ; FVW2-NEXT: [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <2 x i64> [[STEP_ADD7]]
1599 ; FVW2-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, i32* [[TRIGGER]], <2 x i64> [[STEP_ADD8]]
1600 ; FVW2-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <2 x i32> @llvm.masked.gather.v2i32.v2p0i32(<2 x i32*> [[TMP0]], i32 4, <2 x i1> <i1 true, i1 true>, <2 x i32> undef)
1601 ; FVW2-NEXT: [[WIDE_MASKED_GATHER10:%.*]] = call <2 x i32> @llvm.masked.gather.v2i32.v2p0i32(<2 x i32*> [[TMP1]], i32 4, <2 x i1> <i1 true, i1 true>, <2 x i32> undef)
1602 ; FVW2-NEXT: [[WIDE_MASKED_GATHER11:%.*]] = call <2 x i32> @llvm.masked.gather.v2i32.v2p0i32(<2 x i32*> [[TMP2]], i32 4, <2 x i1> <i1 true, i1 true>, <2 x i32> undef)
1603 ; FVW2-NEXT: [[WIDE_MASKED_GATHER12:%.*]] = call <2 x i32> @llvm.masked.gather.v2i32.v2p0i32(<2 x i32*> [[TMP3]], i32 4, <2 x i1> <i1 true, i1 true>, <2 x i32> undef)
1604 ; FVW2-NEXT: [[TMP4:%.*]] = icmp sgt <2 x i32> [[WIDE_MASKED_GATHER]], zeroinitializer
1605 ; FVW2-NEXT: [[TMP5:%.*]] = icmp sgt <2 x i32> [[WIDE_MASKED_GATHER10]], zeroinitializer
1606 ; FVW2-NEXT: [[TMP6:%.*]] = icmp sgt <2 x i32> [[WIDE_MASKED_GATHER11]], zeroinitializer
1607 ; FVW2-NEXT: [[TMP7:%.*]] = icmp sgt <2 x i32> [[WIDE_MASKED_GATHER12]], zeroinitializer
1608 ; FVW2-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT_IN:%.*]], %struct.In* [[IN:%.*]], <2 x i64> [[VEC_IND]], i32 1
1609 ; FVW2-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <2 x i64> [[STEP_ADD]], i32 1
1610 ; FVW2-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <2 x i64> [[STEP_ADD7]], i32 1
1611 ; FVW2-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_IN]], %struct.In* [[IN]], <2 x i64> [[STEP_ADD8]], i32 1
1612 ; FVW2-NEXT: [[WIDE_MASKED_GATHER13:%.*]] = call <2 x float> @llvm.masked.gather.v2f32.v2p0f32(<2 x float*> [[TMP8]], i32 4, <2 x i1> [[TMP4]], <2 x float> undef)
1613 ; FVW2-NEXT: [[WIDE_MASKED_GATHER14:%.*]] = call <2 x float> @llvm.masked.gather.v2f32.v2p0f32(<2 x float*> [[TMP9]], i32 4, <2 x i1> [[TMP5]], <2 x float> undef)
1614 ; FVW2-NEXT: [[WIDE_MASKED_GATHER15:%.*]] = call <2 x float> @llvm.masked.gather.v2f32.v2p0f32(<2 x float*> [[TMP10]], i32 4, <2 x i1> [[TMP6]], <2 x float> undef)
1615 ; FVW2-NEXT: [[WIDE_MASKED_GATHER16:%.*]] = call <2 x float> @llvm.masked.gather.v2f32.v2p0f32(<2 x float*> [[TMP11]], i32 4, <2 x i1> [[TMP7]], <2 x float> undef)
1616 ; FVW2-NEXT: [[TMP12:%.*]] = fadd <2 x float> [[WIDE_MASKED_GATHER13]], <float 5.000000e-01, float 5.000000e-01>
1617 ; FVW2-NEXT: [[TMP13:%.*]] = fadd <2 x float> [[WIDE_MASKED_GATHER14]], <float 5.000000e-01, float 5.000000e-01>
1618 ; FVW2-NEXT: [[TMP14:%.*]] = fadd <2 x float> [[WIDE_MASKED_GATHER15]], <float 5.000000e-01, float 5.000000e-01>
1619 ; FVW2-NEXT: [[TMP15:%.*]] = fadd <2 x float> [[WIDE_MASKED_GATHER16]], <float 5.000000e-01, float 5.000000e-01>
1620 ; FVW2-NEXT: [[TMP16:%.*]] = extractelement <2 x i1> [[TMP4]], i32 0
1621 ; FVW2-NEXT: br i1 [[TMP16]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
1622 ; FVW2: pred.store.if:
1623 ; FVW2-NEXT: [[TMP17:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT:%.*]], i64 [[OFFSET_IDX]]
1624 ; FVW2-NEXT: [[TMP18:%.*]] = extractelement <2 x float> [[TMP12]], i32 0
1625 ; FVW2-NEXT: store float [[TMP18]], float addrspace(1)* [[TMP17]], align 4
1626 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE]]
1627 ; FVW2: pred.store.continue:
1628 ; FVW2-NEXT: [[TMP19:%.*]] = extractelement <2 x i1> [[TMP4]], i32 1
1629 ; FVW2-NEXT: br i1 [[TMP19]], label [[PRED_STORE_IF17:%.*]], label [[PRED_STORE_CONTINUE18:%.*]]
1630 ; FVW2: pred.store.if17:
1631 ; FVW2-NEXT: [[TMP20:%.*]] = or i64 [[OFFSET_IDX]], 16
1632 ; FVW2-NEXT: [[TMP21:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], i64 [[TMP20]]
1633 ; FVW2-NEXT: [[TMP22:%.*]] = extractelement <2 x float> [[TMP12]], i32 1
1634 ; FVW2-NEXT: store float [[TMP22]], float addrspace(1)* [[TMP21]], align 4
1635 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE18]]
1636 ; FVW2: pred.store.continue18:
1637 ; FVW2-NEXT: [[TMP23:%.*]] = extractelement <2 x i1> [[TMP5]], i32 0
1638 ; FVW2-NEXT: br i1 [[TMP23]], label [[PRED_STORE_IF19:%.*]], label [[PRED_STORE_CONTINUE20:%.*]]
1639 ; FVW2: pred.store.if19:
1640 ; FVW2-NEXT: [[TMP24:%.*]] = or i64 [[OFFSET_IDX]], 32
1641 ; FVW2-NEXT: [[TMP25:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], i64 [[TMP24]]
1642 ; FVW2-NEXT: [[TMP26:%.*]] = extractelement <2 x float> [[TMP13]], i32 0
1643 ; FVW2-NEXT: store float [[TMP26]], float addrspace(1)* [[TMP25]], align 4
1644 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE20]]
1645 ; FVW2: pred.store.continue20:
1646 ; FVW2-NEXT: [[TMP27:%.*]] = extractelement <2 x i1> [[TMP5]], i32 1
1647 ; FVW2-NEXT: br i1 [[TMP27]], label [[PRED_STORE_IF21:%.*]], label [[PRED_STORE_CONTINUE22:%.*]]
1648 ; FVW2: pred.store.if21:
1649 ; FVW2-NEXT: [[TMP28:%.*]] = or i64 [[OFFSET_IDX]], 48
1650 ; FVW2-NEXT: [[TMP29:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], i64 [[TMP28]]
1651 ; FVW2-NEXT: [[TMP30:%.*]] = extractelement <2 x float> [[TMP13]], i32 1
1652 ; FVW2-NEXT: store float [[TMP30]], float addrspace(1)* [[TMP29]], align 4
1653 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE22]]
1654 ; FVW2: pred.store.continue22:
1655 ; FVW2-NEXT: [[TMP31:%.*]] = extractelement <2 x i1> [[TMP6]], i32 0
1656 ; FVW2-NEXT: br i1 [[TMP31]], label [[PRED_STORE_IF23:%.*]], label [[PRED_STORE_CONTINUE24:%.*]]
1657 ; FVW2: pred.store.if23:
1658 ; FVW2-NEXT: [[TMP32:%.*]] = or i64 [[OFFSET_IDX]], 64
1659 ; FVW2-NEXT: [[TMP33:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], i64 [[TMP32]]
1660 ; FVW2-NEXT: [[TMP34:%.*]] = extractelement <2 x float> [[TMP14]], i32 0
1661 ; FVW2-NEXT: store float [[TMP34]], float addrspace(1)* [[TMP33]], align 4
1662 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE24]]
1663 ; FVW2: pred.store.continue24:
1664 ; FVW2-NEXT: [[TMP35:%.*]] = extractelement <2 x i1> [[TMP6]], i32 1
1665 ; FVW2-NEXT: br i1 [[TMP35]], label [[PRED_STORE_IF25:%.*]], label [[PRED_STORE_CONTINUE26:%.*]]
1666 ; FVW2: pred.store.if25:
1667 ; FVW2-NEXT: [[TMP36:%.*]] = or i64 [[OFFSET_IDX]], 80
1668 ; FVW2-NEXT: [[TMP37:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], i64 [[TMP36]]
1669 ; FVW2-NEXT: [[TMP38:%.*]] = extractelement <2 x float> [[TMP14]], i32 1
1670 ; FVW2-NEXT: store float [[TMP38]], float addrspace(1)* [[TMP37]], align 4
1671 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE26]]
1672 ; FVW2: pred.store.continue26:
1673 ; FVW2-NEXT: [[TMP39:%.*]] = extractelement <2 x i1> [[TMP7]], i32 0
1674 ; FVW2-NEXT: br i1 [[TMP39]], label [[PRED_STORE_IF27:%.*]], label [[PRED_STORE_CONTINUE28:%.*]]
1675 ; FVW2: pred.store.if27:
1676 ; FVW2-NEXT: [[TMP40:%.*]] = or i64 [[OFFSET_IDX]], 96
1677 ; FVW2-NEXT: [[TMP41:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], i64 [[TMP40]]
1678 ; FVW2-NEXT: [[TMP42:%.*]] = extractelement <2 x float> [[TMP15]], i32 0
1679 ; FVW2-NEXT: store float [[TMP42]], float addrspace(1)* [[TMP41]], align 4
1680 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE28]]
1681 ; FVW2: pred.store.continue28:
1682 ; FVW2-NEXT: [[TMP43:%.*]] = extractelement <2 x i1> [[TMP7]], i32 1
1683 ; FVW2-NEXT: br i1 [[TMP43]], label [[PRED_STORE_IF29:%.*]], label [[PRED_STORE_CONTINUE30]]
1684 ; FVW2: pred.store.if29:
1685 ; FVW2-NEXT: [[TMP44:%.*]] = or i64 [[OFFSET_IDX]], 112
1686 ; FVW2-NEXT: [[TMP45:%.*]] = getelementptr inbounds float, float addrspace(1)* [[OUT]], i64 [[TMP44]]
1687 ; FVW2-NEXT: [[TMP46:%.*]] = extractelement <2 x float> [[TMP15]], i32 1
1688 ; FVW2-NEXT: store float [[TMP46]], float addrspace(1)* [[TMP45]], align 4
1689 ; FVW2-NEXT: br label [[PRED_STORE_CONTINUE30]]
1690 ; FVW2: pred.store.continue30:
1691 ; FVW2-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX6]], 8
1692 ; FVW2-NEXT: [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], <i64 128, i64 128>
1693 ; FVW2-NEXT: [[TMP47:%.*]] = icmp eq i64 [[INDEX_NEXT]], 256
1694 ; FVW2-NEXT: br i1 [[TMP47]], label [[FOR_END:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
1696 ; FVW2-NEXT: ret void
1699 %in.addr = alloca %struct.In addrspace(0)*, align 8
1700 %out.addr = alloca float addrspace(1)*, align 8
1701 %trigger.addr = alloca i32*, align 8
1702 %index.addr = alloca i32*, align 8
1703 %i = alloca i32, align 4
1704 store %struct.In addrspace(0)* %in, %struct.In addrspace(0)** %in.addr, align 8
1705 store float addrspace(1)* %out, float addrspace(1)** %out.addr, align 8
1706 store i32* %trigger, i32** %trigger.addr, align 8
1707 store i32* %index, i32** %index.addr, align 8
1708 store i32 0, i32* %i, align 4
1711 for.cond: ; preds = %for.inc, %entry
1712 %0 = load i32, i32* %i, align 4
1713 %cmp = icmp slt i32 %0, 4096
1714 br i1 %cmp, label %for.body, label %for.end
1716 for.body: ; preds = %for.cond
1717 %1 = load i32, i32* %i, align 4
1718 %idxprom = sext i32 %1 to i64
1719 %2 = load i32*, i32** %trigger.addr, align 8
1720 %arrayidx = getelementptr inbounds i32, i32* %2, i64 %idxprom
1721 %3 = load i32, i32* %arrayidx, align 4
1722 %cmp1 = icmp sgt i32 %3, 0
1723 br i1 %cmp1, label %if.then, label %if.end
1725 if.then: ; preds = %for.body
1726 %4 = load i32, i32* %i, align 4
1727 %idxprom2 = sext i32 %4 to i64
1728 %5 = load %struct.In addrspace(0)*, %struct.In addrspace(0)** %in.addr, align 8
1729 %arrayidx3 = getelementptr inbounds %struct.In, %struct.In addrspace(0)* %5, i64 %idxprom2
1730 %b = getelementptr inbounds %struct.In, %struct.In addrspace(0)* %arrayidx3, i32 0, i32 1
1731 %6 = load float, float addrspace(0)* %b, align 4
1732 %add = fadd float %6, 5.000000e-01
1733 %7 = load i32, i32* %i, align 4
1734 %idxprom4 = sext i32 %7 to i64
1735 %8 = load float addrspace(1)*, float addrspace(1)** %out.addr, align 8
1736 %arrayidx5 = getelementptr inbounds float, float addrspace(1)* %8, i64 %idxprom4
1737 store float %add, float addrspace(1)* %arrayidx5, align 4
1740 if.end: ; preds = %if.then, %for.body
1743 for.inc: ; preds = %if.end
1744 %9 = load i32, i32* %i, align 4
1745 %inc = add nsw i32 %9, 16
1746 store i32 %inc, i32* %i, align 4
1749 for.end: ; preds = %for.cond
1753 ; Using gathers is not profitable for this function. PR48429.
1754 define void @test_gather_not_profitable_pr48429(i32 %d, float* readonly %ptr, float* nocapture %dest) {
1755 ; AVX512-LABEL: @test_gather_not_profitable_pr48429(
1756 ; AVX512-NEXT: entry:
1757 ; AVX512-NEXT: [[IDX_EXT:%.*]] = sext i32 [[D:%.*]] to i64
1758 ; AVX512-NEXT: [[ADD_PTR:%.*]] = getelementptr inbounds float, float* [[PTR:%.*]], i64 [[IDX_EXT]]
1759 ; AVX512-NEXT: [[CMP_NOT10:%.*]] = icmp eq i32 [[D]], 0
1760 ; AVX512-NEXT: br i1 [[CMP_NOT10]], label [[FOR_END:%.*]], label [[FOR_BODY_LR_PH:%.*]]
1761 ; AVX512: for.body.lr.ph:
1762 ; AVX512-NEXT: [[MUL:%.*]] = sub nsw i32 0, [[D]]
1763 ; AVX512-NEXT: [[IDXPROM:%.*]] = sext i32 [[MUL]] to i64
1764 ; AVX512-NEXT: [[TMP0:%.*]] = shl nsw i64 [[IDX_EXT]], 2
1765 ; AVX512-NEXT: [[TMP1:%.*]] = add nsw i64 [[TMP0]], -4
1766 ; AVX512-NEXT: [[TMP2:%.*]] = lshr exact i64 [[TMP1]], 2
1767 ; AVX512-NEXT: [[TMP3:%.*]] = add nuw nsw i64 [[TMP2]], 1
1768 ; AVX512-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP1]], 60
1769 ; AVX512-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[FOR_BODY_PREHEADER:%.*]], label [[VECTOR_MEMCHECK:%.*]]
1770 ; AVX512: vector.memcheck:
1771 ; AVX512-NEXT: [[TMP4:%.*]] = shl nsw i64 [[IDX_EXT]], 2
1772 ; AVX512-NEXT: [[TMP5:%.*]] = add nsw i64 [[TMP4]], -4
1773 ; AVX512-NEXT: [[TMP6:%.*]] = lshr exact i64 [[TMP5]], 2
1774 ; AVX512-NEXT: [[TMP7:%.*]] = shl nsw i64 [[TMP5]], 2
1775 ; AVX512-NEXT: [[TMP8:%.*]] = or i64 [[TMP7]], 2
1776 ; AVX512-NEXT: [[SCEVGEP:%.*]] = getelementptr float, float* [[DEST:%.*]], i64 [[TMP8]]
1777 ; AVX512-NEXT: [[TMP9:%.*]] = add nuw nsw i64 [[TMP6]], 1
1778 ; AVX512-NEXT: [[SCEVGEP4:%.*]] = getelementptr float, float* [[PTR]], i64 [[TMP9]]
1779 ; AVX512-NEXT: [[SCEVGEP6:%.*]] = getelementptr float, float* [[PTR]], i64 [[IDXPROM]]
1780 ; AVX512-NEXT: [[TMP10:%.*]] = add nuw nsw i64 [[TMP6]], 1
1781 ; AVX512-NEXT: [[TMP11:%.*]] = sub i64 [[TMP10]], [[IDX_EXT]]
1782 ; AVX512-NEXT: [[SCEVGEP8:%.*]] = getelementptr float, float* [[PTR]], i64 [[TMP11]]
1783 ; AVX512-NEXT: [[BOUND0:%.*]] = icmp ugt float* [[SCEVGEP4]], [[DEST]]
1784 ; AVX512-NEXT: [[BOUND1:%.*]] = icmp ugt float* [[SCEVGEP]], [[PTR]]
1785 ; AVX512-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
1786 ; AVX512-NEXT: [[BOUND010:%.*]] = icmp ugt float* [[SCEVGEP8]], [[DEST]]
1787 ; AVX512-NEXT: [[BOUND111:%.*]] = icmp ult float* [[SCEVGEP6]], [[SCEVGEP]]
1788 ; AVX512-NEXT: [[FOUND_CONFLICT12:%.*]] = and i1 [[BOUND010]], [[BOUND111]]
1789 ; AVX512-NEXT: [[CONFLICT_RDX:%.*]] = or i1 [[FOUND_CONFLICT]], [[FOUND_CONFLICT12]]
1790 ; AVX512-NEXT: br i1 [[CONFLICT_RDX]], label [[FOR_BODY_PREHEADER]], label [[VECTOR_PH:%.*]]
1791 ; AVX512: vector.ph:
1792 ; AVX512-NEXT: [[N_VEC:%.*]] = and i64 [[TMP3]], 9223372036854775792
1793 ; AVX512-NEXT: [[IND_END:%.*]] = getelementptr float, float* [[PTR]], i64 [[N_VEC]]
1794 ; AVX512-NEXT: [[TMP12:%.*]] = shl i64 [[N_VEC]], 4
1795 ; AVX512-NEXT: [[IND_END14:%.*]] = getelementptr float, float* [[DEST]], i64 [[TMP12]]
1796 ; AVX512-NEXT: [[TMP13:%.*]] = add nsw i64 [[N_VEC]], -16
1797 ; AVX512-NEXT: [[TMP14:%.*]] = lshr exact i64 [[TMP13]], 4
1798 ; AVX512-NEXT: [[TMP15:%.*]] = add nuw nsw i64 [[TMP14]], 1
1799 ; AVX512-NEXT: [[XTRAITER:%.*]] = and i64 [[TMP15]], 7
1800 ; AVX512-NEXT: [[TMP16:%.*]] = icmp ult i64 [[TMP13]], 112
1801 ; AVX512-NEXT: br i1 [[TMP16]], label [[MIDDLE_BLOCK_UNR_LCSSA:%.*]], label [[VECTOR_PH_NEW:%.*]]
1802 ; AVX512: vector.ph.new:
1803 ; AVX512-NEXT: [[UNROLL_ITER:%.*]] = and i64 [[TMP15]], 2305843009213693944
1804 ; AVX512-NEXT: br label [[VECTOR_BODY:%.*]]
1805 ; AVX512: vector.body:
1806 ; AVX512-NEXT: [[POINTER_PHI:%.*]] = phi float* [ [[DEST]], [[VECTOR_PH_NEW]] ], [ [[PTR_IND_7:%.*]], [[VECTOR_BODY]] ]
1807 ; AVX512-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH_NEW]] ], [ [[INDEX_NEXT_7:%.*]], [[VECTOR_BODY]] ]
1808 ; AVX512-NEXT: [[NITER:%.*]] = phi i64 [ [[UNROLL_ITER]], [[VECTOR_PH_NEW]] ], [ [[NITER_NSUB_7:%.*]], [[VECTOR_BODY]] ]
1809 ; AVX512-NEXT: [[NEXT_GEP:%.*]] = getelementptr float, float* [[PTR]], i64 [[INDEX]]
1810 ; AVX512-NEXT: [[TMP17:%.*]] = getelementptr float, float* [[POINTER_PHI]], <16 x i64> <i64 0, i64 16, i64 32, i64 48, i64 64, i64 80, i64 96, i64 112, i64 128, i64 144, i64 160, i64 176, i64 192, i64 208, i64 224, i64 240>
1811 ; AVX512-NEXT: [[TMP18:%.*]] = getelementptr inbounds float, float* [[NEXT_GEP]], i64 [[IDXPROM]]
1812 ; AVX512-NEXT: [[TMP19:%.*]] = bitcast float* [[TMP18]] to <16 x float>*
1813 ; AVX512-NEXT: [[WIDE_LOAD:%.*]] = load <16 x float>, <16 x float>* [[TMP19]], align 4, !alias.scope !2
1814 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[WIDE_LOAD]], <16 x float*> [[TMP17]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope !5, !noalias !7
1815 ; AVX512-NEXT: [[TMP20:%.*]] = bitcast float* [[NEXT_GEP]] to <16 x float>*
1816 ; AVX512-NEXT: [[WIDE_LOAD15:%.*]] = load <16 x float>, <16 x float>* [[TMP20]], align 4, !alias.scope !9
1817 ; AVX512-NEXT: [[TMP21:%.*]] = getelementptr inbounds float, <16 x float*> [[TMP17]], i64 1
1818 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[WIDE_LOAD15]], <16 x float*> [[TMP21]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope !5, !noalias !7
1819 ; AVX512-NEXT: [[INDEX_NEXT:%.*]] = or i64 [[INDEX]], 16
1820 ; AVX512-NEXT: [[PTR_IND:%.*]] = getelementptr float, float* [[POINTER_PHI]], i64 256
1821 ; AVX512-NEXT: [[NEXT_GEP_1:%.*]] = getelementptr float, float* [[PTR]], i64 [[INDEX_NEXT]]
1822 ; AVX512-NEXT: [[TMP22:%.*]] = getelementptr float, float* [[PTR_IND]], <16 x i64> <i64 0, i64 16, i64 32, i64 48, i64 64, i64 80, i64 96, i64 112, i64 128, i64 144, i64 160, i64 176, i64 192, i64 208, i64 224, i64 240>
1823 ; AVX512-NEXT: [[TMP23:%.*]] = getelementptr inbounds float, float* [[NEXT_GEP_1]], i64 [[IDXPROM]]
1824 ; AVX512-NEXT: [[TMP24:%.*]] = bitcast float* [[TMP23]] to <16 x float>*
1825 ; AVX512-NEXT: [[WIDE_LOAD_1:%.*]] = load <16 x float>, <16 x float>* [[TMP24]], align 4, !alias.scope !2
1826 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[WIDE_LOAD_1]], <16 x float*> [[TMP22]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope !5, !noalias !7
1827 ; AVX512-NEXT: [[TMP25:%.*]] = bitcast float* [[NEXT_GEP_1]] to <16 x float>*
1828 ; AVX512-NEXT: [[WIDE_LOAD15_1:%.*]] = load <16 x float>, <16 x float>* [[TMP25]], align 4, !alias.scope !9
1829 ; AVX512-NEXT: [[TMP26:%.*]] = getelementptr inbounds float, <16 x float*> [[TMP22]], i64 1
1830 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[WIDE_LOAD15_1]], <16 x float*> [[TMP26]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope !5, !noalias !7
1831 ; AVX512-NEXT: [[INDEX_NEXT_1:%.*]] = or i64 [[INDEX]], 32
1832 ; AVX512-NEXT: [[PTR_IND_1:%.*]] = getelementptr float, float* [[POINTER_PHI]], i64 512
1833 ; AVX512-NEXT: [[NEXT_GEP_2:%.*]] = getelementptr float, float* [[PTR]], i64 [[INDEX_NEXT_1]]
1834 ; AVX512-NEXT: [[TMP27:%.*]] = getelementptr float, float* [[PTR_IND_1]], <16 x i64> <i64 0, i64 16, i64 32, i64 48, i64 64, i64 80, i64 96, i64 112, i64 128, i64 144, i64 160, i64 176, i64 192, i64 208, i64 224, i64 240>
1835 ; AVX512-NEXT: [[TMP28:%.*]] = getelementptr inbounds float, float* [[NEXT_GEP_2]], i64 [[IDXPROM]]
1836 ; AVX512-NEXT: [[TMP29:%.*]] = bitcast float* [[TMP28]] to <16 x float>*
1837 ; AVX512-NEXT: [[WIDE_LOAD_2:%.*]] = load <16 x float>, <16 x float>* [[TMP29]], align 4, !alias.scope !2
1838 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[WIDE_LOAD_2]], <16 x float*> [[TMP27]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope !5, !noalias !7
1839 ; AVX512-NEXT: [[TMP30:%.*]] = bitcast float* [[NEXT_GEP_2]] to <16 x float>*
1840 ; AVX512-NEXT: [[WIDE_LOAD15_2:%.*]] = load <16 x float>, <16 x float>* [[TMP30]], align 4, !alias.scope !9
1841 ; AVX512-NEXT: [[TMP31:%.*]] = getelementptr inbounds float, <16 x float*> [[TMP27]], i64 1
1842 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[WIDE_LOAD15_2]], <16 x float*> [[TMP31]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope !5, !noalias !7
1843 ; AVX512-NEXT: [[INDEX_NEXT_2:%.*]] = or i64 [[INDEX]], 48
1844 ; AVX512-NEXT: [[PTR_IND_2:%.*]] = getelementptr float, float* [[POINTER_PHI]], i64 768
1845 ; AVX512-NEXT: [[NEXT_GEP_3:%.*]] = getelementptr float, float* [[PTR]], i64 [[INDEX_NEXT_2]]
1846 ; AVX512-NEXT: [[TMP32:%.*]] = getelementptr float, float* [[PTR_IND_2]], <16 x i64> <i64 0, i64 16, i64 32, i64 48, i64 64, i64 80, i64 96, i64 112, i64 128, i64 144, i64 160, i64 176, i64 192, i64 208, i64 224, i64 240>
1847 ; AVX512-NEXT: [[TMP33:%.*]] = getelementptr inbounds float, float* [[NEXT_GEP_3]], i64 [[IDXPROM]]
1848 ; AVX512-NEXT: [[TMP34:%.*]] = bitcast float* [[TMP33]] to <16 x float>*
1849 ; AVX512-NEXT: [[WIDE_LOAD_3:%.*]] = load <16 x float>, <16 x float>* [[TMP34]], align 4, !alias.scope !2
1850 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[WIDE_LOAD_3]], <16 x float*> [[TMP32]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope !5, !noalias !7
1851 ; AVX512-NEXT: [[TMP35:%.*]] = bitcast float* [[NEXT_GEP_3]] to <16 x float>*
1852 ; AVX512-NEXT: [[WIDE_LOAD15_3:%.*]] = load <16 x float>, <16 x float>* [[TMP35]], align 4, !alias.scope !9
1853 ; AVX512-NEXT: [[TMP36:%.*]] = getelementptr inbounds float, <16 x float*> [[TMP32]], i64 1
1854 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[WIDE_LOAD15_3]], <16 x float*> [[TMP36]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope !5, !noalias !7
1855 ; AVX512-NEXT: [[INDEX_NEXT_3:%.*]] = or i64 [[INDEX]], 64
1856 ; AVX512-NEXT: [[PTR_IND_3:%.*]] = getelementptr float, float* [[POINTER_PHI]], i64 1024
1857 ; AVX512-NEXT: [[NEXT_GEP_4:%.*]] = getelementptr float, float* [[PTR]], i64 [[INDEX_NEXT_3]]
1858 ; AVX512-NEXT: [[TMP37:%.*]] = getelementptr float, float* [[PTR_IND_3]], <16 x i64> <i64 0, i64 16, i64 32, i64 48, i64 64, i64 80, i64 96, i64 112, i64 128, i64 144, i64 160, i64 176, i64 192, i64 208, i64 224, i64 240>
1859 ; AVX512-NEXT: [[TMP38:%.*]] = getelementptr inbounds float, float* [[NEXT_GEP_4]], i64 [[IDXPROM]]
1860 ; AVX512-NEXT: [[TMP39:%.*]] = bitcast float* [[TMP38]] to <16 x float>*
1861 ; AVX512-NEXT: [[WIDE_LOAD_4:%.*]] = load <16 x float>, <16 x float>* [[TMP39]], align 4, !alias.scope !2
1862 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[WIDE_LOAD_4]], <16 x float*> [[TMP37]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope !5, !noalias !7
1863 ; AVX512-NEXT: [[TMP40:%.*]] = bitcast float* [[NEXT_GEP_4]] to <16 x float>*
1864 ; AVX512-NEXT: [[WIDE_LOAD15_4:%.*]] = load <16 x float>, <16 x float>* [[TMP40]], align 4, !alias.scope !9
1865 ; AVX512-NEXT: [[TMP41:%.*]] = getelementptr inbounds float, <16 x float*> [[TMP37]], i64 1
1866 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[WIDE_LOAD15_4]], <16 x float*> [[TMP41]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope !5, !noalias !7
1867 ; AVX512-NEXT: [[INDEX_NEXT_4:%.*]] = or i64 [[INDEX]], 80
1868 ; AVX512-NEXT: [[PTR_IND_4:%.*]] = getelementptr float, float* [[POINTER_PHI]], i64 1280
1869 ; AVX512-NEXT: [[NEXT_GEP_5:%.*]] = getelementptr float, float* [[PTR]], i64 [[INDEX_NEXT_4]]
1870 ; AVX512-NEXT: [[TMP42:%.*]] = getelementptr float, float* [[PTR_IND_4]], <16 x i64> <i64 0, i64 16, i64 32, i64 48, i64 64, i64 80, i64 96, i64 112, i64 128, i64 144, i64 160, i64 176, i64 192, i64 208, i64 224, i64 240>
1871 ; AVX512-NEXT: [[TMP43:%.*]] = getelementptr inbounds float, float* [[NEXT_GEP_5]], i64 [[IDXPROM]]
1872 ; AVX512-NEXT: [[TMP44:%.*]] = bitcast float* [[TMP43]] to <16 x float>*
1873 ; AVX512-NEXT: [[WIDE_LOAD_5:%.*]] = load <16 x float>, <16 x float>* [[TMP44]], align 4, !alias.scope !2
1874 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[WIDE_LOAD_5]], <16 x float*> [[TMP42]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope !5, !noalias !7
1875 ; AVX512-NEXT: [[TMP45:%.*]] = bitcast float* [[NEXT_GEP_5]] to <16 x float>*
1876 ; AVX512-NEXT: [[WIDE_LOAD15_5:%.*]] = load <16 x float>, <16 x float>* [[TMP45]], align 4, !alias.scope !9
1877 ; AVX512-NEXT: [[TMP46:%.*]] = getelementptr inbounds float, <16 x float*> [[TMP42]], i64 1
1878 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[WIDE_LOAD15_5]], <16 x float*> [[TMP46]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope !5, !noalias !7
1879 ; AVX512-NEXT: [[INDEX_NEXT_5:%.*]] = or i64 [[INDEX]], 96
1880 ; AVX512-NEXT: [[PTR_IND_5:%.*]] = getelementptr float, float* [[POINTER_PHI]], i64 1536
1881 ; AVX512-NEXT: [[NEXT_GEP_6:%.*]] = getelementptr float, float* [[PTR]], i64 [[INDEX_NEXT_5]]
1882 ; AVX512-NEXT: [[TMP47:%.*]] = getelementptr float, float* [[PTR_IND_5]], <16 x i64> <i64 0, i64 16, i64 32, i64 48, i64 64, i64 80, i64 96, i64 112, i64 128, i64 144, i64 160, i64 176, i64 192, i64 208, i64 224, i64 240>
1883 ; AVX512-NEXT: [[TMP48:%.*]] = getelementptr inbounds float, float* [[NEXT_GEP_6]], i64 [[IDXPROM]]
1884 ; AVX512-NEXT: [[TMP49:%.*]] = bitcast float* [[TMP48]] to <16 x float>*
1885 ; AVX512-NEXT: [[WIDE_LOAD_6:%.*]] = load <16 x float>, <16 x float>* [[TMP49]], align 4, !alias.scope !2
1886 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[WIDE_LOAD_6]], <16 x float*> [[TMP47]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope !5, !noalias !7
1887 ; AVX512-NEXT: [[TMP50:%.*]] = bitcast float* [[NEXT_GEP_6]] to <16 x float>*
1888 ; AVX512-NEXT: [[WIDE_LOAD15_6:%.*]] = load <16 x float>, <16 x float>* [[TMP50]], align 4, !alias.scope !9
1889 ; AVX512-NEXT: [[TMP51:%.*]] = getelementptr inbounds float, <16 x float*> [[TMP47]], i64 1
1890 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[WIDE_LOAD15_6]], <16 x float*> [[TMP51]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope !5, !noalias !7
1891 ; AVX512-NEXT: [[INDEX_NEXT_6:%.*]] = or i64 [[INDEX]], 112
1892 ; AVX512-NEXT: [[PTR_IND_6:%.*]] = getelementptr float, float* [[POINTER_PHI]], i64 1792
1893 ; AVX512-NEXT: [[NEXT_GEP_7:%.*]] = getelementptr float, float* [[PTR]], i64 [[INDEX_NEXT_6]]
1894 ; AVX512-NEXT: [[TMP52:%.*]] = getelementptr float, float* [[PTR_IND_6]], <16 x i64> <i64 0, i64 16, i64 32, i64 48, i64 64, i64 80, i64 96, i64 112, i64 128, i64 144, i64 160, i64 176, i64 192, i64 208, i64 224, i64 240>
1895 ; AVX512-NEXT: [[TMP53:%.*]] = getelementptr inbounds float, float* [[NEXT_GEP_7]], i64 [[IDXPROM]]
1896 ; AVX512-NEXT: [[TMP54:%.*]] = bitcast float* [[TMP53]] to <16 x float>*
1897 ; AVX512-NEXT: [[WIDE_LOAD_7:%.*]] = load <16 x float>, <16 x float>* [[TMP54]], align 4, !alias.scope !2
1898 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[WIDE_LOAD_7]], <16 x float*> [[TMP52]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope !5, !noalias !7
1899 ; AVX512-NEXT: [[TMP55:%.*]] = bitcast float* [[NEXT_GEP_7]] to <16 x float>*
1900 ; AVX512-NEXT: [[WIDE_LOAD15_7:%.*]] = load <16 x float>, <16 x float>* [[TMP55]], align 4, !alias.scope !9
1901 ; AVX512-NEXT: [[TMP56:%.*]] = getelementptr inbounds float, <16 x float*> [[TMP52]], i64 1
1902 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[WIDE_LOAD15_7]], <16 x float*> [[TMP56]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope !5, !noalias !7
1903 ; AVX512-NEXT: [[INDEX_NEXT_7]] = add nuw i64 [[INDEX]], 128
1904 ; AVX512-NEXT: [[PTR_IND_7]] = getelementptr float, float* [[POINTER_PHI]], i64 2048
1905 ; AVX512-NEXT: [[NITER_NSUB_7]] = add i64 [[NITER]], -8
1906 ; AVX512-NEXT: [[NITER_NCMP_7:%.*]] = icmp eq i64 [[NITER_NSUB_7]], 0
1907 ; AVX512-NEXT: br i1 [[NITER_NCMP_7]], label [[MIDDLE_BLOCK_UNR_LCSSA]], label [[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
1908 ; AVX512: middle.block.unr-lcssa:
1909 ; AVX512-NEXT: [[POINTER_PHI_UNR:%.*]] = phi float* [ [[DEST]], [[VECTOR_PH]] ], [ [[PTR_IND_7]], [[VECTOR_BODY]] ]
1910 ; AVX512-NEXT: [[INDEX_UNR:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT_7]], [[VECTOR_BODY]] ]
1911 ; AVX512-NEXT: [[LCMP_MOD_NOT:%.*]] = icmp eq i64 [[XTRAITER]], 0
1912 ; AVX512-NEXT: br i1 [[LCMP_MOD_NOT]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY_EPIL:%.*]]
1913 ; AVX512: vector.body.epil:
1914 ; AVX512-NEXT: [[POINTER_PHI_EPIL:%.*]] = phi float* [ [[PTR_IND_EPIL:%.*]], [[VECTOR_BODY_EPIL]] ], [ [[POINTER_PHI_UNR]], [[MIDDLE_BLOCK_UNR_LCSSA]] ]
1915 ; AVX512-NEXT: [[INDEX_EPIL:%.*]] = phi i64 [ [[INDEX_NEXT_EPIL:%.*]], [[VECTOR_BODY_EPIL]] ], [ [[INDEX_UNR]], [[MIDDLE_BLOCK_UNR_LCSSA]] ]
1916 ; AVX512-NEXT: [[EPIL_ITER:%.*]] = phi i64 [ [[EPIL_ITER_SUB:%.*]], [[VECTOR_BODY_EPIL]] ], [ [[XTRAITER]], [[MIDDLE_BLOCK_UNR_LCSSA]] ]
1917 ; AVX512-NEXT: [[NEXT_GEP_EPIL:%.*]] = getelementptr float, float* [[PTR]], i64 [[INDEX_EPIL]]
1918 ; AVX512-NEXT: [[TMP57:%.*]] = getelementptr float, float* [[POINTER_PHI_EPIL]], <16 x i64> <i64 0, i64 16, i64 32, i64 48, i64 64, i64 80, i64 96, i64 112, i64 128, i64 144, i64 160, i64 176, i64 192, i64 208, i64 224, i64 240>
1919 ; AVX512-NEXT: [[TMP58:%.*]] = getelementptr inbounds float, float* [[NEXT_GEP_EPIL]], i64 [[IDXPROM]]
1920 ; AVX512-NEXT: [[TMP59:%.*]] = bitcast float* [[TMP58]] to <16 x float>*
1921 ; AVX512-NEXT: [[WIDE_LOAD_EPIL:%.*]] = load <16 x float>, <16 x float>* [[TMP59]], align 4, !alias.scope !2
1922 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[WIDE_LOAD_EPIL]], <16 x float*> [[TMP57]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope !5, !noalias !7
1923 ; AVX512-NEXT: [[TMP60:%.*]] = bitcast float* [[NEXT_GEP_EPIL]] to <16 x float>*
1924 ; AVX512-NEXT: [[WIDE_LOAD15_EPIL:%.*]] = load <16 x float>, <16 x float>* [[TMP60]], align 4, !alias.scope !9
1925 ; AVX512-NEXT: [[TMP61:%.*]] = getelementptr inbounds float, <16 x float*> [[TMP57]], i64 1
1926 ; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[WIDE_LOAD15_EPIL]], <16 x float*> [[TMP61]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope !5, !noalias !7
1927 ; AVX512-NEXT: [[INDEX_NEXT_EPIL]] = add nuw i64 [[INDEX_EPIL]], 16
1928 ; AVX512-NEXT: [[PTR_IND_EPIL]] = getelementptr float, float* [[POINTER_PHI_EPIL]], i64 256
1929 ; AVX512-NEXT: [[EPIL_ITER_SUB]] = add i64 [[EPIL_ITER]], -1
1930 ; AVX512-NEXT: [[EPIL_ITER_CMP_NOT:%.*]] = icmp eq i64 [[EPIL_ITER_SUB]], 0
1931 ; AVX512-NEXT: br i1 [[EPIL_ITER_CMP_NOT]], label [[MIDDLE_BLOCK]], label [[VECTOR_BODY_EPIL]], !llvm.loop [[LOOP11:![0-9]+]]
1932 ; AVX512: middle.block:
1933 ; AVX512-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC]]
1934 ; AVX512-NEXT: br i1 [[CMP_N]], label [[FOR_END]], label [[FOR_BODY_PREHEADER]]
1935 ; AVX512: for.body.preheader:
1936 ; AVX512-NEXT: [[PTR_ADDR_012_PH:%.*]] = phi float* [ [[PTR]], [[VECTOR_MEMCHECK]] ], [ [[PTR]], [[FOR_BODY_LR_PH]] ], [ [[IND_END]], [[MIDDLE_BLOCK]] ]
1937 ; AVX512-NEXT: [[DEST_ADDR_011_PH:%.*]] = phi float* [ [[DEST]], [[VECTOR_MEMCHECK]] ], [ [[DEST]], [[FOR_BODY_LR_PH]] ], [ [[IND_END14]], [[MIDDLE_BLOCK]] ]
1938 ; AVX512-NEXT: br label [[FOR_BODY:%.*]]
1940 ; AVX512-NEXT: [[PTR_ADDR_012:%.*]] = phi float* [ [[INCDEC_PTR:%.*]], [[FOR_BODY]] ], [ [[PTR_ADDR_012_PH]], [[FOR_BODY_PREHEADER]] ]
1941 ; AVX512-NEXT: [[DEST_ADDR_011:%.*]] = phi float* [ [[ADD_PTR6:%.*]], [[FOR_BODY]] ], [ [[DEST_ADDR_011_PH]], [[FOR_BODY_PREHEADER]] ]
1942 ; AVX512-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[PTR_ADDR_012]], i64 [[IDXPROM]]
1943 ; AVX512-NEXT: [[TMP62:%.*]] = load float, float* [[ARRAYIDX]], align 4
1944 ; AVX512-NEXT: store float [[TMP62]], float* [[DEST_ADDR_011]], align 4
1945 ; AVX512-NEXT: [[TMP63:%.*]] = load float, float* [[PTR_ADDR_012]], align 4
1946 ; AVX512-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds float, float* [[DEST_ADDR_011]], i64 1
1947 ; AVX512-NEXT: store float [[TMP63]], float* [[ARRAYIDX5]], align 4
1948 ; AVX512-NEXT: [[INCDEC_PTR]] = getelementptr inbounds float, float* [[PTR_ADDR_012]], i64 1
1949 ; AVX512-NEXT: [[ADD_PTR6]] = getelementptr inbounds float, float* [[DEST_ADDR_011]], i64 16
1950 ; AVX512-NEXT: [[CMP_NOT:%.*]] = icmp eq float* [[INCDEC_PTR]], [[ADD_PTR]]
1951 ; AVX512-NEXT: br i1 [[CMP_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP13:![0-9]+]]
1953 ; AVX512-NEXT: ret void
1955 ; FVW2-LABEL: @test_gather_not_profitable_pr48429(
1957 ; FVW2-NEXT: [[IDX_EXT:%.*]] = sext i32 [[D:%.*]] to i64
1958 ; FVW2-NEXT: [[ADD_PTR:%.*]] = getelementptr inbounds float, float* [[PTR:%.*]], i64 [[IDX_EXT]]
1959 ; FVW2-NEXT: [[CMP_NOT10:%.*]] = icmp eq i32 [[D]], 0
1960 ; FVW2-NEXT: br i1 [[CMP_NOT10]], label [[FOR_END:%.*]], label [[FOR_BODY_LR_PH:%.*]]
1961 ; FVW2: for.body.lr.ph:
1962 ; FVW2-NEXT: [[MUL:%.*]] = sub nsw i32 0, [[D]]
1963 ; FVW2-NEXT: [[IDXPROM:%.*]] = sext i32 [[MUL]] to i64
1964 ; FVW2-NEXT: [[TMP0:%.*]] = shl nsw i64 [[IDX_EXT]], 2
1965 ; FVW2-NEXT: [[TMP1:%.*]] = add nsw i64 [[TMP0]], -4
1966 ; FVW2-NEXT: [[TMP2:%.*]] = lshr exact i64 [[TMP1]], 2
1967 ; FVW2-NEXT: [[TMP3:%.*]] = add nuw nsw i64 [[TMP2]], 1
1968 ; FVW2-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP1]], 12
1969 ; FVW2-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[FOR_BODY_PREHEADER:%.*]], label [[VECTOR_MEMCHECK:%.*]]
1970 ; FVW2: vector.memcheck:
1971 ; FVW2-NEXT: [[TMP4:%.*]] = shl nsw i64 [[IDX_EXT]], 2
1972 ; FVW2-NEXT: [[TMP5:%.*]] = add nsw i64 [[TMP4]], -4
1973 ; FVW2-NEXT: [[TMP6:%.*]] = lshr exact i64 [[TMP5]], 2
1974 ; FVW2-NEXT: [[TMP7:%.*]] = shl nsw i64 [[TMP5]], 2
1975 ; FVW2-NEXT: [[TMP8:%.*]] = or i64 [[TMP7]], 2
1976 ; FVW2-NEXT: [[SCEVGEP:%.*]] = getelementptr float, float* [[DEST:%.*]], i64 [[TMP8]]
1977 ; FVW2-NEXT: [[TMP9:%.*]] = add nuw nsw i64 [[TMP6]], 1
1978 ; FVW2-NEXT: [[SCEVGEP4:%.*]] = getelementptr float, float* [[PTR]], i64 [[TMP9]]
1979 ; FVW2-NEXT: [[SCEVGEP6:%.*]] = getelementptr float, float* [[PTR]], i64 [[IDXPROM]]
1980 ; FVW2-NEXT: [[TMP10:%.*]] = add nuw nsw i64 [[TMP6]], 1
1981 ; FVW2-NEXT: [[TMP11:%.*]] = sub i64 [[TMP10]], [[IDX_EXT]]
1982 ; FVW2-NEXT: [[SCEVGEP8:%.*]] = getelementptr float, float* [[PTR]], i64 [[TMP11]]
1983 ; FVW2-NEXT: [[BOUND0:%.*]] = icmp ugt float* [[SCEVGEP4]], [[DEST]]
1984 ; FVW2-NEXT: [[BOUND1:%.*]] = icmp ugt float* [[SCEVGEP]], [[PTR]]
1985 ; FVW2-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
1986 ; FVW2-NEXT: [[BOUND010:%.*]] = icmp ugt float* [[SCEVGEP8]], [[DEST]]
1987 ; FVW2-NEXT: [[BOUND111:%.*]] = icmp ult float* [[SCEVGEP6]], [[SCEVGEP]]
1988 ; FVW2-NEXT: [[FOUND_CONFLICT12:%.*]] = and i1 [[BOUND010]], [[BOUND111]]
1989 ; FVW2-NEXT: [[CONFLICT_RDX:%.*]] = or i1 [[FOUND_CONFLICT]], [[FOUND_CONFLICT12]]
1990 ; FVW2-NEXT: br i1 [[CONFLICT_RDX]], label [[FOR_BODY_PREHEADER]], label [[VECTOR_PH:%.*]]
1992 ; FVW2-NEXT: [[N_VEC:%.*]] = and i64 [[TMP3]], 9223372036854775804
1993 ; FVW2-NEXT: [[IND_END:%.*]] = getelementptr float, float* [[PTR]], i64 [[N_VEC]]
1994 ; FVW2-NEXT: [[TMP12:%.*]] = shl i64 [[N_VEC]], 4
1995 ; FVW2-NEXT: [[IND_END14:%.*]] = getelementptr float, float* [[DEST]], i64 [[TMP12]]
1996 ; FVW2-NEXT: br label [[VECTOR_BODY:%.*]]
1997 ; FVW2: vector.body:
1998 ; FVW2-NEXT: [[POINTER_PHI:%.*]] = phi float* [ [[DEST]], [[VECTOR_PH]] ], [ [[PTR_IND:%.*]], [[VECTOR_BODY]] ]
1999 ; FVW2-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
2000 ; FVW2-NEXT: [[NEXT_GEP:%.*]] = getelementptr float, float* [[PTR]], i64 [[INDEX]]
2001 ; FVW2-NEXT: [[TMP13:%.*]] = getelementptr float, float* [[POINTER_PHI]], <2 x i64> <i64 0, i64 16>
2002 ; FVW2-NEXT: [[TMP14:%.*]] = getelementptr float, float* [[POINTER_PHI]], <2 x i64> <i64 32, i64 48>
2003 ; FVW2-NEXT: [[TMP15:%.*]] = getelementptr inbounds float, float* [[NEXT_GEP]], i64 [[IDXPROM]]
2004 ; FVW2-NEXT: [[TMP16:%.*]] = bitcast float* [[TMP15]] to <2 x float>*
2005 ; FVW2-NEXT: [[WIDE_LOAD:%.*]] = load <2 x float>, <2 x float>* [[TMP16]], align 4, !alias.scope !7
2006 ; FVW2-NEXT: [[TMP17:%.*]] = getelementptr inbounds float, float* [[TMP15]], i64 2
2007 ; FVW2-NEXT: [[TMP18:%.*]] = bitcast float* [[TMP17]] to <2 x float>*
2008 ; FVW2-NEXT: [[WIDE_LOAD16:%.*]] = load <2 x float>, <2 x float>* [[TMP18]], align 4, !alias.scope !7
2009 ; FVW2-NEXT: call void @llvm.masked.scatter.v2f32.v2p0f32(<2 x float> [[WIDE_LOAD]], <2 x float*> [[TMP13]], i32 4, <2 x i1> <i1 true, i1 true>), !alias.scope !10, !noalias !12
2010 ; FVW2-NEXT: call void @llvm.masked.scatter.v2f32.v2p0f32(<2 x float> [[WIDE_LOAD16]], <2 x float*> [[TMP14]], i32 4, <2 x i1> <i1 true, i1 true>), !alias.scope !10, !noalias !12
2011 ; FVW2-NEXT: [[TMP19:%.*]] = bitcast float* [[NEXT_GEP]] to <2 x float>*
2012 ; FVW2-NEXT: [[WIDE_LOAD17:%.*]] = load <2 x float>, <2 x float>* [[TMP19]], align 4, !alias.scope !14
2013 ; FVW2-NEXT: [[TMP20:%.*]] = getelementptr float, float* [[NEXT_GEP]], i64 2
2014 ; FVW2-NEXT: [[TMP21:%.*]] = bitcast float* [[TMP20]] to <2 x float>*
2015 ; FVW2-NEXT: [[WIDE_LOAD18:%.*]] = load <2 x float>, <2 x float>* [[TMP21]], align 4, !alias.scope !14
2016 ; FVW2-NEXT: [[TMP22:%.*]] = getelementptr inbounds float, <2 x float*> [[TMP13]], i64 1
2017 ; FVW2-NEXT: [[TMP23:%.*]] = getelementptr inbounds float, <2 x float*> [[TMP14]], i64 1
2018 ; FVW2-NEXT: call void @llvm.masked.scatter.v2f32.v2p0f32(<2 x float> [[WIDE_LOAD17]], <2 x float*> [[TMP22]], i32 4, <2 x i1> <i1 true, i1 true>), !alias.scope !10, !noalias !12
2019 ; FVW2-NEXT: call void @llvm.masked.scatter.v2f32.v2p0f32(<2 x float> [[WIDE_LOAD18]], <2 x float*> [[TMP23]], i32 4, <2 x i1> <i1 true, i1 true>), !alias.scope !10, !noalias !12
2020 ; FVW2-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
2021 ; FVW2-NEXT: [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
2022 ; FVW2-NEXT: [[PTR_IND]] = getelementptr float, float* [[POINTER_PHI]], i64 64
2023 ; FVW2-NEXT: br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]]
2024 ; FVW2: middle.block:
2025 ; FVW2-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC]]
2026 ; FVW2-NEXT: br i1 [[CMP_N]], label [[FOR_END]], label [[FOR_BODY_PREHEADER]]
2027 ; FVW2: for.body.preheader:
2028 ; FVW2-NEXT: [[PTR_ADDR_012_PH:%.*]] = phi float* [ [[PTR]], [[VECTOR_MEMCHECK]] ], [ [[PTR]], [[FOR_BODY_LR_PH]] ], [ [[IND_END]], [[MIDDLE_BLOCK]] ]
2029 ; FVW2-NEXT: [[DEST_ADDR_011_PH:%.*]] = phi float* [ [[DEST]], [[VECTOR_MEMCHECK]] ], [ [[DEST]], [[FOR_BODY_LR_PH]] ], [ [[IND_END14]], [[MIDDLE_BLOCK]] ]
2030 ; FVW2-NEXT: br label [[FOR_BODY:%.*]]
2032 ; FVW2-NEXT: [[PTR_ADDR_012:%.*]] = phi float* [ [[INCDEC_PTR:%.*]], [[FOR_BODY]] ], [ [[PTR_ADDR_012_PH]], [[FOR_BODY_PREHEADER]] ]
2033 ; FVW2-NEXT: [[DEST_ADDR_011:%.*]] = phi float* [ [[ADD_PTR6:%.*]], [[FOR_BODY]] ], [ [[DEST_ADDR_011_PH]], [[FOR_BODY_PREHEADER]] ]
2034 ; FVW2-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[PTR_ADDR_012]], i64 [[IDXPROM]]
2035 ; FVW2-NEXT: [[TMP25:%.*]] = load float, float* [[ARRAYIDX]], align 4
2036 ; FVW2-NEXT: store float [[TMP25]], float* [[DEST_ADDR_011]], align 4
2037 ; FVW2-NEXT: [[TMP26:%.*]] = load float, float* [[PTR_ADDR_012]], align 4
2038 ; FVW2-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds float, float* [[DEST_ADDR_011]], i64 1
2039 ; FVW2-NEXT: store float [[TMP26]], float* [[ARRAYIDX5]], align 4
2040 ; FVW2-NEXT: [[INCDEC_PTR]] = getelementptr inbounds float, float* [[PTR_ADDR_012]], i64 1
2041 ; FVW2-NEXT: [[ADD_PTR6]] = getelementptr inbounds float, float* [[DEST_ADDR_011]], i64 16
2042 ; FVW2-NEXT: [[CMP_NOT:%.*]] = icmp eq float* [[INCDEC_PTR]], [[ADD_PTR]]
2043 ; FVW2-NEXT: br i1 [[CMP_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]]
2045 ; FVW2-NEXT: ret void
2048 %idx.ext = sext i32 %d to i64
2049 %add.ptr = getelementptr inbounds float, float* %ptr, i64 %idx.ext
2050 %cmp.not10 = icmp eq i32 %d, 0
2051 br i1 %cmp.not10, label %for.end, label %for.body.lr.ph
2053 for.body.lr.ph: ; preds = %entry
2054 %mul = sub nsw i32 0, %d
2055 %idxprom = sext i32 %mul to i64
2058 for.body: ; preds = %for.body.lr.ph, %for.body
2059 %ptr.addr.012 = phi float* [ %ptr, %for.body.lr.ph ], [ %incdec.ptr, %for.body ]
2060 %dest.addr.011 = phi float* [ %dest, %for.body.lr.ph ], [ %add.ptr6, %for.body ]
2061 %arrayidx = getelementptr inbounds float, float* %ptr.addr.012, i64 %idxprom
2062 %0 = load float, float* %arrayidx, align 4
2063 store float %0, float* %dest.addr.011, align 4
2064 %1 = load float, float* %ptr.addr.012, align 4
2065 %arrayidx5 = getelementptr inbounds float, float* %dest.addr.011, i64 1
2066 store float %1, float* %arrayidx5, align 4
2067 %incdec.ptr = getelementptr inbounds float, float* %ptr.addr.012, i64 1
2068 %add.ptr6 = getelementptr inbounds float, float* %dest.addr.011, i64 16
2069 %cmp.not = icmp eq float* %incdec.ptr, %add.ptr
2070 br i1 %cmp.not, label %for.end.loopexit, label %for.body
2072 for.end.loopexit: ; preds = %for.body
2075 for.end: ; preds = %for.end.loopexit, %entry